Method of stratifying haematuria patients

Abstract

Identifying the causes of haematuria requires detailed clinical investigations. Unbalanced patient populations (mostly male) and diverse causes complicate patient classification and biomarker identification. Multi-marker analysis with advanced methods can improve diagnosis even in unbalanced datasets. We used k-means clustering, decision trees, random forest with LIME explainer, and CACTUS algorithm to classify patients as healthy (with no clear cause of haematuria) or sick (with an identified cause of haematuria e.g., bladder cancer, or infection). The CACTUS algorithm was more robust to dataset imbalance, achieving balanced accuracy of 0.747 for both genders, 0.718 for females, and 0.803 for males. Key biomarkers identified included microalbumin, male gender, and tPSA for the whole dataset; age, microalbumin, tPSA, cystatin C, BTA, HAD, and S100A4 for males; and microalbumin, IL-8, pERK, and CXCL16 for females. CACTUS outperformed decision trees and random forest, highlighting potential novel biomarkers for future diagnosis.

Description

[0001] Method of Stratifying Haematuria Patients

[0002] Introduction

[0003] Haematuria is defined as the visible presence of red blood cells (RBCs) in urine (gross haematuria) or at least three RBCs in high-powered field upon microscopic evaluation of a urine sample. Prevalence of microhaematuria among the general population is relatively high. It was estimated that in 2.4 -31.1 % of total urine samples, RBCs are detectable in concentrations exceeding a fixed reference threshold (Barocas etal., 2020; Britton etal., 1992; Messing etal., 2006; Mohr et al., 1986). A haematuria patient population can be heterogenous with differences in age, gender, risk factors, geographical diversity etc., and it can have very different aetiology, including the presence of genitourinary malignant diseases. While most commonly cases of haematuria are non-malignant (e.g., infection, kidney or bladder stones, benign prostate enlargement, menstrual blood contamination), first-stage assessment should always be focused on the physical examination and collection of patient history, current treatment (e.g., anticoagulants) (Ingelfinger, 2021), lifestyle (e.g., smoking, alcohol consumption, strenuous physical activity), and occupational hazards. Additionally, the aforementioned factors have been identified as risk factors for urinary tract cancer (Loo et al., 2013). Dipstick urine analysis can be performed to confirm or exclude some causes of haematuria, for example, infection. For non-obvious cases, further investigation should be performed. Currently, cystoscopy together with urine cytology is the gold standard for bladder cancer diagnosis. Cystoscopy is an invasive procedure which is not without risk e.g., infection, bleeding, and pain. The need to engage specialized equipment and trained diagnostic staff may delay the time to diagnosis, which is crucial in all types of cancer. Computed tomography (CT) urography is warranted for patients who require upper urinary tract investigation, which raises concerns of radiation exposure (Nawfel et al., 2004).

[0004] To improve haematuria patient management and reduce the burden of invasive procedures or radiation exposure from imaging procedures, many recommendations and guidelines have been published (Georgieva et al., 2019). Georgieva et al also compared the advantages, harms, and costs associated with different haematuria evaluation guidelines (Georgieva et al., 2019). The analysis showed that guidelines which missed detection for the fewest number of cancers also generated the highest number of radiation-induced cancer cases, 81 false-positive cases, and the highest cost for diagnostic procedures. One of the most interesting findings is that uniform CT imaging for patients is associated with a limited increase in cancer detection, high personal cost and is generally uneconomical. Although guidelines are constantly being updated, they vary considerably and there is still no consensus on the best haematuria patient management solution, not only in terms of imaging procedures but also on the usefulness of analysing biomarkers in the initial screening process. Biomarkers have been used to screen and identify numerous diseases, allowing personalized profiling of patients (Abogunrin et al., 2012; Dorcely et al., 2017). Samples for biomarker analysis are usually easy to obtain (e.g., blood, urine) and do not require invasive and costly procedures. However, due to the heterogeneity of the haematuria patient population, disease pathways and the wide range of potential causes of haematuria, the development of biomarker-based screening or diagnostic tests is a significant challenge. To date, only two biomarkers - nuclear matrix protein (NMP22) and bladder tumour antigen (BTA) - have been approved by the Food and Drug Administration (FDA) for the detection and monitoring of bladder cancer. Unfortunately, commercially available tests for both markers have low specificity and high false-positive rates (Guo et al., 2014; Sharma et al., 1999). Data shows that combining biomarker screening (NMP22) with cytology may improve patient screening (Sajid et al., 2020), but current guidelines do not recommend the use of urinary tumour markers or cytology in the initial evaluation of microhaematuria. To improve the diagnostic pathway, current research has focused on developing a rapid bedside point of care (POC) diagnostic test to differentiate patients with benign and malignant disease (Abogunrin et al., 2012). Despite numerous clinical and biomarker studies developing a POC test has proved challenging and involves high financial costs (Dimashkieh et al., 2013; Dogan et al., 2013; Sutton et al., 2018). Additionally, as more pathological processes can cause haematuria and more biomarkers need to be analysed, the analysis becomes more complex and less tractable, creating a need for computational tools to help generate personalised insights from the available disease data (Mathew and Pillai, 2015; Raghupathi and Raghupathi,2014; Ryu and Song, 2014).

[0005] Numerous studies have proved that using advanced analytical methods, it is possible to create algorithms that can improve patient diagnosis with multiple marker analysis (Abogunrin et al., 2012; Duggan et al., 2022; Emmert-Streib et al., 2013). Machine Learning (ML) (Dwivedi, 2018; Rasheed et al., 2020), especially, has been able to produce unique insights using different data sources (Nambiar et al., 2013; O'Leary, 2013; Sun and Reddy, 2013). Unfortunately, applying ML algorithms is not always a straightforward task, especially in real-world medical data. Traditional ML models have been developed assuming a balanced distribution of classes within a dataset, which can be difficult to achieve with medical data. This is one of the main reasons for the poor generalization of ML algorithms. These models pay equal attention to the majority and minority classes. As a result, they often perform poorly on the minority class, especially when the imbalance in the data is extreme (Zhang et al., 2010). For example, in haematuria cohorts, where one or more classes are underrepresented, there is potential for misclassification.

[0006] In the case of haematuria, the patient's condition is described by different types of information such as lifestyle, imaging data and laboratory tests. This causes two main problems: high data dimensionality and the challenge of analysing different data types (such as categorical and numerical) in one test. Data dimensionality is a major challenge for ML algorithms, especially when dealing with small datasets where the number of features exceeds the number of samples. Non-meaningful parameters need to be separated to subtract hidden information and provide actionable insights to clinicians. This could be achieved at the level of domain experts and data driven features that could be incorporated into the model design. In addition, with high uncertainty in disease development and missing information, it is very challenging to use ML to understand pathways and generate knowledge.

[0007] The final challenge for ML, and currently a requirement for any clinical decision support system, is explainability. Explainability is a property of an Al algorithm that allows a human to understand why a particular decision was made. In practice, explainability can either be an inherent property of an algorithm, or it can be approximated by other methods. Many modern ML methods can outperform humans in certain analytical tasks (e.g., pattern recognition), but they lack explainability, so the explanation must be approximated. On the other hand, the performance of traditional explainable methods is usually inferior to modern state-of-the-art methods such as neural networks, so the tradeoff between performance and explainability is a major challenge for modern clinical decision support systems.

[0008] Given the challenges described above in diagnosing patients with haematuria, we applied the Comprehensive Abstraction and Classification Tool for Uncovering Structures (CACTUS) (Gherardini et al., 2024) algorithm to a Haematuria Biomarker (HaBio) (Duggan et al., 2022) dataset and compared it with other classification methods to facilitate the diagnosis procedure and provide actionable insights for clinical patient management. This resulted in finding novel biomakers and combinations of biomarkers for the stratification of haematuria patients.

[0009] References

[0010] Abogunrin, F., O'Kane, H.F., Ruddock, M.W., Stevenson, M., Reid, C.N., O'Sullivan, J.M., Anderson, N.H., et al. (2012), "The impact of biomarkers in multivariate algorithms for bladder cancer diagnosis in patients with hematuria", Cancer, Vol.118 No. 10, pp. 2641-2650, doi: 10.1002 / cncr.26544.

[0011] Barabasi, A.-L., Gulbahce, N. and Loscalzo, J. (2011), "Network medicine: a network-based approach to human disease", Nature Reviews Genetics, Vol. 12 No. 1, pp. 56-68, doi: 10.1038 / nrg2918.

[0012] Barocas, D.A., Boorjian, S.A., Alvarez, R.D., Downs, T.M., Gross, C.P., Hamilton, B.D., Kobashi, K.C., et al. (2020), "Microhematuria: AUA / SUFU Guideline", Journal of Urology, Wolters Kluwer Health, Vol. 204 No. 4, pp. 778-786, doi:10.1097 / JU.0000000000001297.

[0013] Bossuyt, P.M., Reitsma, J.B., Bruns, D.E., Gatsonis, C.A., Glasziou, P.P., Irwig, L.M., Lijmer, J.G., et al. (2003), "Towards complete and accurate reporting of studies of diagnostic accuracy: the STARD initiative", BMJ, BMJ Publishing Group Ltd, Vol.326 No. 7379, pp. 41-44, doi: 10.1136 / bmj.326.7379.41. Britton, J.P., Dowell, A.C., Whelan, P. and Harris, C.M. (1992), "A community study of bladder cancer screening by the detection of occult urinary bleeding", Journal of Urology, Vol. 148 No. 3 I, pp. 788- 790, doi: 10.1016 / 50022-5347(17)36720-4.

[0014] Bunone, G., Briand, P.-A., Miksicek, RJ. and Picard, D. (1996), Activation of the Unliganded Estrogen Receptor by EGF Involves the MAP Kinase Pathway and Direct Phosphorylation, The EMBO Journal, Vol. 15.

[0015] Chew, J.S., Saleem, M., Florkowski, C.M. and George, P.M. (2008), Cystatin C-A Paradigm of Evidence Based Laboratory Medicine, Clin Biochem Rev, Vol. 29.

[0016] Dimashkieh, H., Wolff, D.J., Smith, T.M., Houser, P.M., Nietert, PJ. and Yang, J. (2013), "Evaluation of urovysion and cytology for bladder cancer detection: A study of 1835 paired urine samples with clinical and histologic correlation", Cancer Cytopathology, Vol. 121 No. 10, pp. 591-597, doi: 10.1002 / cncy.21327.

[0017] Dogan, C., Pelit, E.S., Yildinm, A., Zemheri, I.E., anakci, C., Ba§ok, E.K. and a§kurlu, T. (2013), "The value of the NMP22 test for superficial bladder cancer diagnosis and follow-up", Turkish Journal of Urology, Turkish Association of Urology, Vol. 39 No. 3, pp. 137-142, doi: 10.5152 / tud.2013.029.

[0018] Dorcely, B., Katz, K., Jagannathan, R., Chiang, S.S., Oluwadare, B., Goldberg, LJ. and Bergman, M. (2017), "Novel biomarkers for prediabetes, diabetes, and associated complications", Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy, Dove Medical Press, Vol. 10, pp. 345-361, doi: 10.2147 / DMSO.S100074.

[0019] Duggan, B., O'Rourke, D., Anderson, N., Reid, C.N., Watt, J., O'Kane, H., Boyd, R., et al. (2022), "Biomarkers to assess the risk of bladder cancer in patients presenting with haematuria are genderspecific", Frontiers in Oncology, Frontiers Media S.A.,Vol. 12, doi: 10.3389 / fonc.2022.1009014.

[0020] Dwivedi, A.K. (2018), "Analysis of computational intelligence techniques for diabetes mellitus prediction", Neural Computing and Applications, Vol. 30 No. 12, pp. 3837-3845, doi: 10.1007 / s00521- 017-2969-9.

[0021] Elewa, U., Sanchez-Nino, M.D., Mahillo-Fernandez, I., Martin-Cleary, C., Belen Sanz, A., Perez-Gomez, M.V., Fernandez-Fernandez, B., et al. (2016), "Circulating CXCL16 in Diabetic Kidney Disease", Kidney and Blood Pressure Research, S. Karger AG, Vol. 41 No. 5, pp. 663-671, doi: 10.1159 / 000447935.

[0022] Emmert-Streib, F., Abogunrin, F., de Matos Simoes, R., Duggan, B., Ruddock, M.W., Reid, C.N., Roddy, O., et al. (2013), "Collectives of diagnostic biomarkers identify high-risk subpopulations of hematuria patients: Exploiting heterogeneity in largescale biomarker data", BMC Medicine, BioMed Central Ltd., Vol. 11 No. 1, doi:10.1186 / 1741-7015-11-12.

[0023] Gao, J., Li, J. and Ma, L. (2005), "Regulation of EGF-induced ERK / MAPK activation and EGFR internalization by G protein-coupled receptor kinase 2", Acta Biochimica et Biophysica Sinica, Vol. 37 No. 8, pp. 525-531, doi: 10.1111 / j.l745-7270.2005.00076.x.

[0024] Georgieva, M. V., Wheeler, S.B., Erim, D., Smith-Bindman, R., Loo, R., Ng, C., Garg, T., et al. (2019), "Comparison of the Harms, Advantages, and Costs Associated with Alternative Guidelines for the Evaluation of Hematuria", JAMA Internal Medicine, American Medical Association, Vol. 179 No. 10, pp. 1352-1362, doi:10.1001 / jamainternmed.2019.2280.

[0025] Gherardini, L., Varma, V.R., Capala, K., Woods, R. and Sousa, J. (2024), "CACTUS: a Comprehensive Abstraction and Classification Tool for Uncovering Structures", ACM Transactions on Intelligent Systems and Technology, doi: 10.1145 / 3649459.

[0026] Guo, A., Wang, X., Gao, L., Shi, J., Sun, C. and Wan, Z. (2014), "Bladder tumour antigen (BTA stat) test compared to the urine cytology in the diagnosis of bladder cancer: A meta-analysis", Journal of the Canadian Urological Association, Canadian Medical Association, Vol. 8 No. 5-6, doi: 10.5489 / cuaj.l668.

[0027] Ingelfinger, J.R. (2021), "Hematuria in Adults", edited by Longo, D.L.New England Journal of Medicine, Massachusetts Medical Society, Vol. 385 No. 2, pp. 153-163, doi: 10.1056 / NEJMral604481.

[0028] Izumi, K., Zheng, Y., Li, Y., Zaengle, J. and Miyamoto, H. (2012), "Epidermal growth factor induces bladder cancer cell proliferation through activation of the androgen receptor", International Journal of Oncology, Vol. 41 No. 5, pp. 1587-1592, doi:10.3892 / ijo.2012.1593.

[0029] Khoury, C.C., Chen, S. and Ziyadeh, F.N. (2019), "Pathophysiology of Diabetic Nephropathy", Chronic Renal Disease, Elsevier, pp. 279-296, doi: 10.1016 / B978-0-12-815876-0.00019-X.

[0030] Ko, Y.-C., Mukaida, N., Ishiyama, S., Tokue, A., Kawai, T., Matsushima, K. and Kasahara4, T. (1993), Elevated lnterleukin-8 Levels in the Urine of Patients with Urinary Tract Infections, INFEcrION AND IMMUNITY, Vol. 61.

[0031] Lang, K., Bonberg, N., Robens, S., Behrens, T., Hovanec, J., Deix, T., Braun, K., et al. (2017), Soluble Chemokine (C-X-C Motif) Ligand 16 (CXCL16) in Urine as a Novel Biomarker Candidate to Identify High Grade and Muscle Invasive Urothelial Carcinomas, Vol. 8.

[0032] Lokeshwar, V.B., Obek, C., Pham, H.T., Wei, D., Young, M.J., Duncan, R.C., Soloway, M.S., et al. (2000), "Urinary hyaluronic acid and hyaluronidase:markers for bladder cancer detection and evaluation of grade", Journal of Urology, Vol. 163 No. 1, pp. 348-356, doi: 10.1016 / 50022-5347(05)68050-0.

[0033] Loo, R.K., Lieberman, S.F., Slezak, J.M., Landa, H.M., Mariani, A.J., Nicolaisen, G.,Aspera, A.M., et al. (2013), "Stratifying risk of urinary tract malignant tumors in patients with asymptomatic microscopic hematuria", Mayo Clinic Proceedings, Elsevier Ltd, Vol. 88 No. 2, pp. 129-138, doi: 10.1016 / j.mayocp.2012.10.004.

[0034] Luo, L., Kieneker, L.M., Van Der Vegt, B., Bakker, S.J.L., Gruppen, E.G., Casteleijn, N.F., De Boer, R.A., et al. (2023), "Urinary albumin excretion and cancer risk: The PREVEND cohort study", Nephrology Dialysis Transplantation, Oxford University Press, Vol. 38 No. 12, pp. 2723-2732, doi: 10.1093 / ndt / gfadl07.

[0035] Mathew, P.S. and Pillai, A.S. (2015), "Big Data solutions in Healthcare: Problems and perspectives", 2015 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS), pp. 1-6, doi:10.1109 / ICIIECS.2015.7193211.

[0036] Messing, E.M., Madeb, R., Young, T., Gilchrist, K.W., Bram, L., Greenberg, E.B., Wegenke, J.D., et al. (2006), "Long-term outcome of hematuria home screening for bladder cancer in men", Cancer, Vol. 107 No. 9, pp. 2173-2179, doi:10.1002 / cncr.22224.

[0037] Mohr, D.N., Offord, K.P., Owen, R.A. and Melton III, LJ. (1986), "Asymptomatic Microhematuria and Urologic Disease: A Population-Based Study", JAMA, Vol. 256 No. 2, pp. 224-229, doi: 10.1001 / jama.1986.03380020086028.

[0038] Murphy, P.M. (2003), "CXC Chemokines", in Henry, H.L. and Norman, A.W. (Eds.), Encyclopedia of Hormones, Academic Press, New York, pp. 351-362, doi: https: / / doi.org / 10.1016 / B0-12-341103- 3 / 00059-0.

[0039] Najafi, M., Ahmadi, A. and Mortezaee, K. (2019), "Extracellular-signal-regulated kinase / mitogen- activated protein kinase signaling as a target for cancer therapy: an updated review", Cell Biology International, Wiley-Blackwell Publishing Ltd, 1 November, doi: 10.1002 / cbin.11187. Nambiar, R., Bhardwaj, R., Sethi, A. and Vargheese, R. (2013), "A Look at Challenges and Opportunities of Big Data Analytics in Healthcare", 2013 IEEE International Conference on Big Data, pp. 17-22, doi: 10.1109 / BigData.2013.6691753.

[0040] Nawfel, R.D., Judy, P.F., Schleipman, A.R. and Silverman, S.G. (2004), "Patient radiation dose at CT urography and conventional urography", Radiology, Vol. 232 No. 1, pp. 126-132, doi: 10.1148 / radiol.2321030222.

[0041] O'Leary, D.E. (2013), "Artificial Intelligence and Big Data", IEEE Intelligent Systems, Vol. 28 No. 2, pp. 96-99, doi: 10.1109 / mis.2013.39.

[0042] Poudel, B., Yadav, B.K., Nepal, A.K., Jha, B. and Raut, K.B. (2012), "Prevalence and association of microalbuminuria in essential hypertensive patients", North American Journal of Medical Sciences, Vol. 4 No. 8, pp. 331-335, doi: 10.4103 / 1947-2714.99501.

[0043] Qiao, L.-Y. and Gulick, M.A. (2007), "Region-specific changes in the phosphorylation of ERK1 / 2 and ERK5 in rat micturition pathways following cyclophosphamide-induced cystitis", Am J Physiol Regul Integr Comp Physiol, Vol. 292, pp. 1368-1375, doi: 10.1152 / ajpregu.00570.2006. -Chronic.

[0044] R Core Team. (2023), "R: A Language and Environment for Statistical Computing", R Foundation for Statistical Computing, Vienna, Austria.

[0045] Raghupathi, W. and Raghupathi, V. (2014), "Big data analytics in healthcare: promise and potential.", Health Information Science and Systems, Vol. 2 No. 1, pp. 3-10, doi: 10.1186 / 2047-2501-2-3 PMID - 25825667.

[0046] Rasheed, J., Jamil, A., Hameed, A.A., Aftab, U., Aftab, J., Shah, S.A. and Draheim, D. (2020), "A Survey on Artificial Intelligence Approaches in Supporting Frontline Workers and Decision Makers for COVID- 19 Pandemic", Chaos, Solitons & Fractals, Vol. 141, p. 110337, doi: 10.1016 / j.chaos.2020.110337 PMID -33071481.

[0047] Reza, H.S., Ali, Z., Tara, H. and Ali, B. (2021), "Age-specific reference ranges of prostate-specific antigen in the elderly of Amirkola: A population-based study", Asian Journal of Urology, Editorial Office of Asian Journal of Urology, Vol. 8 No. 2, pp. 183-188, doi: 10.1016 / j.ajur.2020.03.001.

[0048] Ribeiro, M., Singh, S. and Guestrin, C. (2016), '"Why Should I Trust You?': Explaining the Predictions of Any Classifier", Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations, Association for Computational Linguistics, Stroudsburg, PA, USA,pp. 97-101, doi: 10.18653 / vl / N16-3020.

[0049] Ryu, S. and Song, T.-M. (2014), "Big Data Analysis in Healthcare", Healthcare Informatics Research, Vol. 20 No. 4, p. 247, doi: 10.4258 / hir.2014.20.4.247.

[0050] Sagara, Y., Miyata, Y., Iwata, T., Kanda, S., Tomayoshi, H., Sakai, H. and Kanetake, H. (2010), "Clinical significance and prognostic value of S100A4 and matrix metalloproteinase-14 in patients with organ- confined bladder cancer", Experimental and Therapeutic Medicine, Vol. 1 No. 1, pp. 27-31, doi: 10.3892 / etm_00000005.

[0051] Sajid, M.T., Zafar, M.R., Ahmad, H., Ullah, S., Mirza, Z.L and Shahzad, K. (2020), "Diagnostic accuracy of NMP 22 and urine cytology for detection of transitional cell carcinoma urinary bladder taking cystoscopy as gold standard", Pakistan Journal of Medical Sciences, Professional Medical Publications, Vol. 36 No. 4, pp. 705-710, doi: 10.12669 / pjms.36.4.1638.

[0052] Sharma, S., Zippe, C.D., Pandrangi, L., Nelson, D. and Agarwal, A. (1999), "Exclusion criteria enhance the specificity and positive predictive value of NMP22* and BTA stat", The Journal of Urology, Wolters Kluwer Philadelphia, PA, Vol. 162 No. 1, pp. 53-57. Sun, J. and Reddy, C.K. (2013), "Big data analytics for healthcare", ACM.

[0053] Sutton, A.J., Lamont, J. v, Evans, R.M., Williamson, K., O'Rourke, D., Duggan, B., Sagoo, G.S., et al. (2018), "An early analysis of the cost-effectiveness of a diagnostic classifier for risk stratification of haematuria patients (DCRSHP)compared to flexible cystoscopy in the diagnosis of bladder cancer", PLOS ONE, Vol. 13 No. 8, p. e0202796, doi: 10.1371 / journal.pone.0202796 PMID - 30138462.

[0054] Urquidi, V., Chang, M., Dai, Y., Kim, J., Wolfson, E.D., Goodison, S. and Rosser, CJ. (2012), IL-8 as a Urinary Biomarker for the Detection of Bladder Cancer.

[0055] Van den Bussche, C.J., Heaney, C.D., Kates, M., Hooks, J.J., Baloga, K., Sokoll, L., Rosenthal, D., et al. (2024), "Urinary IL-6 and IL-8 as predictive markers in bladder urothelial carcinoma: A pilot study", Cancer Cytopathology, John Wiley & Sons, Ltd, Vol. 132 No. 1, pp. 50-59, doi: https: / / doi.org / 10.1002 / cncy.22767.

[0056] Yao, H., Rayburn, E.R., Shi, Q„ Gao, L., Hu, W. and Li, H. (2017), "FDA-approved drugs that interfere with laboratory tests: A systematic search of US drug labels", Critical Reviews in Clinical Laboratory Sciences, Taylor and Francis Ltd, 2 January, doi: 10.1080 / 10408363.2016.1191425.

[0057] Description of figures

[0058] Figure 1 - Dataset structure ingested and division into study groups.

[0059] Figure 2 - CACTUS analysis results - The ten most important biomarkers according to the rank values.

[0060] Figure 3 - CACTUS analysis results - Probability of flips for sick and healthy categories in descending order of rank value.

[0061] Summary of the invention

[0062] The present invention is based on the realisation that there are significant differences between the biomarkers required to stratify male and female patients presenting with haematuria. It has been shown that females who present with haematuria do not receive the same diagnostic consideration as men and are not referred in a timely manner for urological consultation, and thus have poorer bladder cancer oncological outcomes. Therefore, younger, and healthier female patients require careful consideration when being referred for cystoscopy or CT urography to reduce procedural risk (e.g. radiation exposure, potential infection) while at the same time providing a fast and efficient diagnostic plan. As the underlying diseases, risk factors and associated harms for the diagnostic procedure and misdiagnosis are different for both genders, it is crucial to stratify patients into the different data subsets. The present invention therefore provides specific panels of biomarkers useful in the stratification of female patients presenting with haematuria. In a first aspect of the invention there is a method for aiding the diagnosis of a female patient presenting with haematuria, said method comprising the steps of (i) determining in a sample previously isolated from said patient, the concentration of one or more biomarkers selected from the list consisting of Phospho-Extracellular Signal Regulated Kinase (pERK), Chemokine (C-X-C motif) Ligand 16 (CXCL16), microalbumin and interleukin 8 (IL-8); (ii) stratifying the patient into "healthy" or "sick" categories, wherein the detection of an elevated concentration of the one or more biomarkers compared to a normal control indicates that the patient should be placed into the "sick" category.

[0063] In a second aspect of the invention there is a solid support material comprising binding molecules attached thereto, said binding molecules having affinity specific for pERK and, separately, CXCL16, microalbumin and IL-8 with the binding molecules for each being in discrete locations on the support material.

[0064] In a third aspect of the invention there is a method for aiding in the diagnosis of bladder cancer in a female patient presenting with haematuria, said method comprising detecting the concentration of one or more biomarkers selected from the list comprising pERK, CXCL16, microalbumin and IL-8 in a sample previously isolated from said patient; wherein an elevated level of one or more of the biomarkers compared to a normal control is indicative of risk of bladder cancer.

[0065] Detailed description

[0066] The present invention is based on the finding that certain biomarkers present in a female patient presenting with haematuria, enable a more accurate stratification to be made compared to the prior art methods of diagnosis on the basis of biomarkers that are used for the diagnosis of both men and women. Identification of particular biomarkers in a sample previously isolated from a female patient is indicative of illness in the female patient and enables them to be classified as sick rather than healthy. It has been surprisingly found that these biomarkers differ significantly between men and women.

[0067] As used herein, the term "biomarker" refers to a molecule present in a biological sample obtained from a patient, the concentration of which in said sample may be indicative of a pathological state. Various biomarkers that have been found to be useful in stratifying patients presenting with haematuria, either alone or in combination with other diagnostic methods, or as complementary biomarkers in combination with other biomarkers, are described herein. Diagnosis may be made on the basis of the level of expression or the concentration of the biomarker in a sample previously isolated from the patient. The biomarkers of the present invention are typically identified in a serum or urine sample from the patient. Preferably, the sample is a urine sample.

[0068] The top biomarkers for stratifying haematuria patients into "sick" and "healthy" groups for each statistical analysis method are shown below. The term "both" refers to biomarkers which were important when the patients were not separated by gender.

[0069] CACTUS

[0070] Both - Gender, serum tPSA, microalbumin, CXCL16, sTNFRl, protein, IL-7, midkine, progranulin.

[0071] Males- Age, microalbumin, ACR, BTA, serum cystatin C, MCP-l, serum tPSA, sTNFR2, D dimer, protein.

[0072] Females - Microalbumin, IL-8, CXCL16, serum PAI-l / tPA, IL-la, pERK, creatinine, sTNFRl, progranulin, midkine.

[0073] Random Forest

[0074] Both - serum tPSA, male, microalbumin, sTNFRl, cystatin C, CXCL16, age, osmolarity, pERK, cystatin B.

[0075] Males - Age, osmolarity, serum HDL, serum S100A4, serum HAD, NGAL, clusterin, progranulin, sTNFRl, cystatin C.

[0076] Females - Microalbumin, pERK, IL-8, CXCL16, ACR, cystatin C, progranulin, serum CEA, serum PAI- 1 / tPA, serum EGF.

[0077] Logistic regression

[0078] Both - Creatinine, CXCL16, cystatin C, IL-7, MCP-l, microalbumin, progranulin, protein, PSA / tPSA, TM, sTNFRl, STNFR2, VEGF, serum HDL.

[0079] Males -Clusterin, cystatin C, EGF, IL-la, IL-6, IL-8, MCP-l, microalbumin, progranulin, sTNFRl, STNFR2.

[0080] Females - CXCL16, IL-8, microalbumin, pERK, protein, VEGF.

[0081] Decision trees

[0082] Males - Cystatin C, age, serum CD 44, ACR, serum IL-4, HAD, TGFbl, pERK.

[0083] Females - Microalbumin, serum S100A4, serum CRP, ACR, serum VEGF, serum Cystatin C, serum EGF.

[0084] Preferably, the panel of biomarkers with which the present invention is concerned for the stratification of female haematuria patients into "sick" and "healthy" groups comprises one or more of pERK, CXCL16, Microalbum in and IL-8, and optionally, one or more further biomarkers selected from BTA, Midkine, PAI-l / tPA, 8OHdG, CEA, CK18, Clusterin, Creatinine, Cystatin B, Cystatin C, d- Dimer, EGF, FAS, HAD, IL-la, IL-lb, IL-4, IL-6, IL-7, MCP-1, MMP9NGAL, MMP9TIMP1, NGAL, NSE, Progranulin, TUP, TGFB1, Thrombomodulin, sTNFRl, TPA, VEGF, Triglycerides, and / or the concentration of albumin / microalbumin / protein and creatinine expressed as an albuminxreatinine ratio (ACR). Preferably one of the one or more further biomarkers is selected from EGF and VEGF. In some embodiments the sample has a concentration of albumin and creatinine expressed as an albuminxreatinine ratio. This may be calculated by measuring the concentration separately of albumin and creatinine. The skilled person will appreciate conventional ways to measure albumin and creatinine concentrations. When the kidneys are functioning properly there is virtually no albumin present in the urine.

[0085] Whilst the panel of biomarkers may be any combination of biomarkers listed under the statistical methods above, preferably they are any combination of the biomarkers listed when the groups were split by gender, preferred combinations for stratifying female haematuria patients include pERK and CXCL16, microalbumin and CXCL16, microalbumin and pERK, microalbumin and IL-8, CXCL16 and IL-8, pERK and IL-8, microalbumin, pERK and IL-8, pERK, CXCL16 and IL-8, pERK, CXCL16 and Microalbumin, and pERK, CXCL16, Microalbumin and IL-8; even more preferably the panel of biomarkers for stratifying female haematuria patients comprises (i) pERK and CXCL16; (ii) pERK, CXCL16, IL-8; (iii) pERK, CXCL16 and Microalbumin; or (iv) pERK, CXCL16, Microalbumin and IL-8. A further embodiment comprises any of the two, three or four biomarker combinations of pERK, CXCL16, microalbumin and IL-8 with an additional biomarker selected from EGF and VEGF.

[0086] When the panels of biomarkers described herein are used to aid in the diagnosis of haematuria patients the concentrations are compared to control values. This can help to stratify the patient into "healthy" or "sick" categories wherein the detection of an elevated concentration of the one or more biomarkers compared to a normal control indicates that the patient should be placed into the "sick" category. Control values or "normal controls" as used herein are derived from the concentration of corresponding biomarkers in a biological sample obtained from an individual or individuals who do not have haematuria or any conditions covered by the definition of "sick" as used herein. These may be derived from normal reference ranges for the biomarkers. Such individual(s) may be, for example, healthy individuals or individuals suffering from diseases other than bladder cancer or that do not have haematuria as a symptom. Alternatively, the control values may correspond to the concentration of each of the biomarkers in a sample obtained from the patient prior to experiencing haematuria. For the avoidance of doubt, the term "corresponding biomarkers" means that concentrations of the same combination of biomarkers that are determined in respect of the patient's sample are also used to determine the control values. For example, if the concentration of pERK and CXCL16 in the patient's sample is determined, then the concentration of pERK and CXCL16 in the control is also known.

[0087] In some embodiments the patient may be presenting with haematuria and / or with an infection. For the avoidance of doubt, the term "haematuria" refers to the presence of red blood cells in the urine. Suitably the infection may be a bacterial or viral infection, preferably a bacterial infection. Suitably, the method may further comprise a step of characterising the patient's infection status. Characterising infection means diagnosing the patient as having infection or being infection free, and, may include identifying the infecting species. Infection may be determined using clinical-based diagnoses based on clinical history, biomarkers, dipstick analysis or UTI multiplex array (e.g. Randox Urinary Track Multiplex Assay). By incorporating an initial infection test the AUCs can be increased and it can also reduce the number of biomarkers used to diagnose the bladder cancer. In the context of the present invention, the term "bladder cancer" is understood to include urothelial carcinoma (UC), transitional cell carcinoma, bladder squamous cell carcinoma and / or bladder adenocarcinoma. In some embodiments the presence of haematuria and / or an infection may further increase the elevated levels of the biomarkers within the panel of biomarkers compared to if haematuria and / or the infection were not present in female bladder cancer patients.

[0088] As used herein the term "sick" in terms of classifying haematuria patients refers to a population which had any of the following possible causes for their haematuria: chronic kidney diseases, infection, other benign diagnosis, bladder cancer, history of bladder cancer or other types of cancer (e.g., prostate cancer, renal cell carcinoma). The term "healthy" as used in terms of classifying haematuria patients refers to a population which included every patient with no identified cause of their haematuria.

[0089] Preferably the biomarkers are in the urinary form i.e. are identified in a urine sample. In a preferred embodiment, the biomarkers within the panel may be identified and their concentrations within the sample determined either sequentially or simultaneously in the sample isolated from the patient. The biomarkers may be identified and their concentrations within the isolated sample may be determined by routine methods, which are known in the art, such as by contacting the sample with a substrate having binding molecules specific for each of the biomarkers included in the panel of biomarkers. Preferably the substrate has at least two binding molecules immobilised thereon, more preferably three, four or more binding molecules, wherein each binding molecule is specific to an individual biomarker and the first probe is specific for pERK and the second probe is specific for CXCL16, Microalbumin or IL-8. As used herein, the term "specific" means that the binding molecule binds only to one of the biomarkers of the invention, with negligible binding to other biomarkers of the invention or to other analytes in the biological sample being analysed. This ensures that the integrity of the diagnostic assay and its result using the biomarkers of the invention is not compromised by additional binding events. The biomarker concentrations may be measured by using methodology based on immuno-detection. As such, the binding molecule is preferably an antibody, such as a polyclonal antibody or a monoclonal antibody. As used herein, the term "antibody" includes any immunoglobulin or immunoglobulin-like molecule or fragment thereof, Fab fragments, ScFv fragments and other antigen binding fragments. The term "polyclonal antibodies" refers to a heterogeneous population of antibodies which recognise multiple epitopes on a target / antigen. The term "monoclonal antibodies" refers to a homogenous population of antibodies (including antibody fragments), which recognise a single epitope on a target / antigen.

[0090] Immuno-detection technology is also readily incorporated into transportable or hand-held devices for use outside of the clinical environment. A quantitative immunoassay such as a Western blot or ELISA can be used to detect the amount of protein biomarkers. A preferred method of analysis comprises using a multianalyte biochip which enables several proteins to be detected and quantified simultaneously. 2D Gel Electrophoresis is also a technique that can be used for multi-analyte analysis.

[0091] In a preferred embodiment, the binding molecules are immobilised on a solid support, ready to be contacted with the patient sample. A preferred solid support material is in the form of a biochip. A biochip is typically a planar substrate that may be, for example, mineral or polymer based, but is preferably ceramic. The solid support may be manufactured according to the method disclosed in, for example, GB-A-2324866 the contents of which is incorporated herein in its entirety.

[0092] The solid supports may be screen printed in accordance with known methods disclosed in, for example, W02017 / 085509. Preferably, the Biochip Array Technology system (BAT) (available from Randox Laboratories Limited) may be used to determine the levels of biomarkers in the sample. More preferably, the Evidence Evolution and Evidence Investigator apparatus (available from Randox Laboratories) may be used.

[0093] The solid support material comprises binding molecules attached thereto, said binding molecules having affinity specific for pERK and, separately, CXCL16, microalbumin and IL-8, with the binding molecules each being in discrete locations on the support material. The solid support material may further comprise, each in discrete locations, one or more binding molecules each having affinity specific for an additional biomarker selected from BTA, Midkine, PAI-l / tPA, 8OHdG, CEA, CK18, Clusterin, Creatinine, Cystatin B, Cystatin C, d-Dimer, EGF, FAS, HAD, IL-la, IL-lb, IL-4, IL-6, IL-7, MCP- 1, MMP9NGAL, MMP9TIMP1, NGAL, NSE, Progranulin, TUP, TGFB1, Thrombomodulin, sTNFRl, TPA, VEGF and Triglycerides, preferably EGF, VEGF, ILla, Creatinine, sTNFRl, progranulin and midkine. For example, the binding molecules attached to the solid support material preferably have affinities to the combinations (i) pERK and CXCL16; (ii) pERK, CXCL16 and IL-8; (iii) pERK, CXCL16 and Microalbumin; or (iv) pERK, CXCL16, microalbumin and IL-8. Preferably the solid support of the invention is used to aid in the diagnosis of haematuria patients, more preferably to stratify female haematuria patients into "sick" and "healthy" groups.

[0094] The present invention also provides kits comprising probes for a panel of biomarkers comprising pERK and, separately, one or more of CXCL16, microalbumin, and IL-8 with the binding molecules each being in discrete locations on the support material. The solid support material may further comprise, each in discrete locations, one or more binding molecules each having affinity specific for an additional biomarker selected from BTA, Midkine, PAI-l / tPA, 8OHdG, CEA, CK18, Clusterin, Creatinine, Cystatin B, Cystatin C, d-Dimer, EGF, FAS, HAD, IL-la, IL-lb, IL-4, IL-6, IL-7, MCP-1, Microalbumin, MMP9NGAL, MMP9TIMP1, NGAL, NSE, Progranulin, TUP, TGFB1, Thrombomodulin, sTNFRl, TPA, VEGF and Triglycerides, preferably EGF, VEGF, ILla, Creatinine, sTNFRl, progranulin and midkine and optionally reagents for the measurement of albumin and creatinine. For example, the panel of biomarkers are preferably combinations selected from (i) pERK and CXCL16; (ii) pERK, CXCL16 and IL-8; (iii) pERK, CXCL16 and Microalbumin; (iv) pERK, CXCL16, microalbumin and IL-8. Such kits can be used to stratify female patients presenting with haematuria according to the first aspect of the current invention or may be used to detect bladder cancer or the risk of bladder cancer in a female patient according to the aspect of the invention described below.

[0095] The invention also provides a method for the detection of or the risk of bladder cancer in a female patient comprising the steps of (i) determining that a female patient does not have an infection; (ii) detecting the presence of one or more biomarkers in a sample isolated from the female patient, wherein said one or more biomarkers are selected from pERK, CXCL16, IL-8 and Microalbumin; (iii) assessing the presence or risk of bladder cancer in the female patient wherein detection of an elevated presence of the biomarkers compared to a normal control indicates the presence or the risk of cancer in the female patient from whom the sample is isolated. Suitably, the one or more biomarkers are (i) pERK and CXCL16; (ii) pERK, CXCL16 and IL-8; (iii) pERK, CXCL16 and Microalbumin; or (iv) pERK, CXCL16, microalbumin and IL-8.

[0096] The present invention also provides the use of substrates described herein in a method for the detection of or the risk of bladder cancer in a female patient. In the methods of the present invention, in order for bladder cancer or the risk of bladder cancer to be diagnosed, the biomarkers within the panel of biomarkers tested may be found at an elevated level compared to the corresponding biomarker in a normal control sample. In some embodiments, the concentrations of biomarkers are found at a significantly higher level than in a control sample. The determination of "higher concentration" is relative and determined with respect to control values.

[0097] In a preferred embodiment, each of the female patient and control biomarker concentration values is inputted into one or more statistical algorithms to produce an output value that indicates whether the patient should be placed into the "sick" or "healthy" categories or in some embodiments, if bladder cancer is present in the patient. If the output value is less than the biomarker cut-off, the patient is placed into the "healthy" risk category or is deemed to be negative by biochip for bladder cancer. If the output value is higher than the biomarker cut-off, the patient is placed into the "sick" risk category or deemed to be positive by biochip for bladder cancer. Patients in the "sick" risk category may then be subject to further stratification methods or interventions to determine the cause of haematuria. When two or more biomarkers are to be used in the diagnostic method a suitable mathematical or machine learning classification model, such as logistic regression equation, can be derived. Such models as described herein may be referred to as "statistical methodologies". The significance of the levels of the biomarkers can be established by inputting into said model. Such a classification model may be chosen from at least one of decision trees, artificial neural networks, logistic regression, random forests, support vector machine, clustering algorithms such as CACTUS (categorical clustering using summaries), or indeed any other method developing classification models known in the art. Variables can be logarithmically, or square root transformed in a regression model when data is not normally distributed.

[0098] In one embodiment of the current invention, a Clinical Risk Score (CRS) is calculated for the female patient, which is a cumulative score using, but not restricted to, the following clinical and demographic measurements: age, haematuria (non-visible vs. macro haematuria), smoking (pack years), BMI, blood pressure (controlled, normotensive, hypertensive), occupational risk score (FINJEM), social class (ONS Codes), comorbidities e.g. diabetes, chronic kidney disease (CKD) etc., medications e.g. statins, antihypertensives etc, specific medications (found to increase risk of bladder cancer), pain relief, renal transplant, kidney cancer, other cancers, pelvic radiotherapy and UTIs (with / without microbiology).

[0099] Example scores used when calculating the CRS for a patient: age is greater than 65 equals a score of 1; age is less than 65 equals a score of 0; non-visible haematuria (NVH) equals a score of 1; macro haematuria equals a score of 2. Therefore, a patient who is older than 65 years with macro haematuria would have a cumulative score of 3, using age and haematuria as clinical risk scores.

[0100] In one embodiment, the biomarker test data and CRS are combined to determine if the patient was in the "sick" or "healthy" categories. This information would allow the GP to manage their patients in primary care and refer them to further tests if and when appropriate. For example, patients who present with haematuria and are negative by biomarkers (i.e. the biomarker combinations of the current invention) and have a low CRS would be monitored in primary care by their GP, rather than being referred to have a cystoscopy. Patients who are negative by biomarkers and have a moderate CRS would be referred to urology for cystoscopy (non-urgent). Patients who are positive by biomarkers and have a low CRS would be referred to urology for cystoscopy (non-urgent). Patients who are positive by biomarkers and have moderate CRS would be 'red flagged' for an urgent cystoscopy.

[0101] The accuracy of statistical methods used in accordance with the present invention can be best described by their receiver operating characteristics (ROC). The ROC curve addresses both the sensitivity, the number of true positives, and the specificity, the number of true negatives, of the test. Therefore, sensitivity and specificity values for a given combination of biomarkers are an indication of the accuracy of the assay. For example, if a biomarker combination has sensitivity and specificity values of 80%, out of 100 patients which have bladder cancer, 80 will be correctly identified from the determination of the presence of the particular combination of biomarkers as positive for bladder cancer, while out of 100 patients who have not got bladder cancer 80 will accurately test negative for the disease.

[0102] The ROC also provides a measure of the predictive power of the test in the form of the area under the curve (AUC). AUC is a measure of the probability that the perceived measurement will allow correct identification of a condition. By convention, this area is always > 0.5. Values range between 1.0 (perfect separation of the test values of the two groups) and 0.5 (no apparent distributional difference between the two groups of test values). The area does not depend only on a particular portion of the plot such as the point closest to the diagonal or the sensitivity at 90% specificity, but on the entire plot. This is a quantitative, descriptive expression of how close the ROC plot is to the perfect one (area = 1.0).

[0103] As a general rule, a test with a sensitivity of about 80% or more and a specificity of about 80% or more is regarded in the art as a test of potential use, although these values vary according to the clinical application. In a preferred embodiment, the panel of biomarkers has an AUC value of at least 0.7, preferably at least 0.75. It is well understood in the art that biomarker normal or 'background' concentrations may exhibit slight variation due to, for example, age, gender or ethnic / geographical genotypes. As a result, the cut-off value used in the methods of the invention may also slightly vary due to optimization depending upon the target patient or population. Adjusting the cut-off will also allow the operator to increase the sensitivity at the expense of specificity and vice versa. In one embodiment, the algorithm has a sensitivity and / or specificity of at least 0.7 respectively. Preferably, the algorithm has a sensitivity of at least 0.75. Where two or more biomarkers are used in the invention, a suitable mathematical or machine learning classification model, such as logistic regression equation, can be derived. The skilled statistician will understand how such a suitable model is derived, which can include other variables such as age and gender of the patient. The ROC curve can be used to assess the accuracy of the model, and the model can be used independently or in an algorithm to aid clinical decision making. Although a logistic regression equation is a common mathematical / statistical procedure used in such cases and an option in the context of the present invention, other mathematical / statistical, decision trees or machine learning procedures can also be used. The skilled person will appreciate that the model generated for a given population may need to be adjusted for application to datasets obtained from different populations or patient cohorts.

[0104] Biomarkers which were found to be important in stratifying females are the phosphorylated form of ERK and epidermal growth factor (EGF). ERKs are members of the mitogen-activated protein kinase (MARK) family, which is activated by a sequential phosphorylation cascade. Of this family, ERK1 and ERK2 are the two most studied kinases. When phosphorylated, ERK1 and ERK2 translocate to the nucleus, where they phosphorylate transcription factors involved in cell cycle regulation and tissue proliferation. MARK signalling is active in both early and advanced stages of tumourigenesis and promotes tumour proliferation, survival, and metastasis (Najafi et al., 2019). In addition, urinary expression of pERK has been shown to be increased in cyclophosphamide-induced cystitis in an animal model (Qiao and Gulick, 2007). EGF has also been shown to activate the MAPK / ERK pathway (Bunone et al., 1996; Gao et al., 2005). EGF, acting through the EGF receptor, promotes cancer development (Yin et al., 2022). EGF has been shown to promote bladder cancer cell proliferation by modulating androgen receptor signalling, which plays an important role in the progression of bladder cancer (Izumi et al., 2012). To the best of our knowledge, this is the first time that EGF has been described as a potential biomarker for stratifying haematuria patients or the detection of the pathology related to urinary tract cancer, providing an initial estimate of the possible concentration of the biomarker for decision making.

[0105] Microalbumin has been described in the literature as a marker of renal dysfunction, most commonly in diabetic nephropathy (Khoury et al., 2019) and hypertension (Poudel et al., 2012). There is some evidence that elevated levels of microalbumin may be associated with some types of cancer, including cancer of the urinary tract (Luo et al., 2023). In the literature, values of microalbumin below 20 mg / mL are considered physiologically normal, but according to our analysis, the decision boundaries were much lower. For both genders the decision boundaries for microalbumin were 7.9 mg / mL (CACTUS and decision trees) and 10.23 mg / mL (random forest). For the male subset of data, the boundary for CACTUS was 13.43 mg / mL, while for the female dataset the boundary was 7.9 mg / mL for CACTUS and decision trees and 5.38 mg / mL for random forest. The values above the decision boundaries are classified as important for the stratification process and are more indicative of sick individuals. Therefore, when using the official reference values, it is possible to miss some individuals with developing pathology. It is important to note that microalbumin is one of the biomarkers that is independently important for almost all stratification and applied models.

[0106] Several biomarkers were common to more than one group including CXCL16, cystatin C and microalbumin (described above). CXCL16 is a cholesterol receptor and a chemokine with a potential role in vascular injury, angiogenesis, and inflammation. CXCL16 has previously been described to be elevated in patients with urothelial cancer (Lang et al., 2017; Murphy, 2003). It has also been shown to be associated with diabetic kidney disease (Elewa et al., 2016). As CXCL16 is not a routinely studied biomarker, reference values for it have not yet been described, but according to our studies, elevated levels are associated with the pathological causes of underlying haematuria. Urinary levels of CXCL16 above 0.3 ng / mL (CACTUS) and 0.2 ng / mL (random forest) regardless of gender and above 0.1 ng / mL (random forest) and 0.2 ng / mL (CACTUS) when considering female patients, may be of use in the stratification of patients presenting with haematuria.

[0107] For the female stratification process, many of the most important characteristics differ from the male and both genders datasets. One of the selected biomarkers is interleukin-8 (IL-8) measured in urine. IL-8 is an angiogenic factor associated with inflammation and carcinogenesis. It has been shown that elevated urinary levels of IL-8 are associated with urothelial cell carcinoma (Ko et al., 1993; Van den Bussche et al., 2024). In the study of Urquidi et al. (Urquidi et al., 2012) it has been shown that the urinary level of IL-8 in patients is elevated when compared to healthy controls with the median value of 128.43 pg / ml vs. 0 pg / ml, respectively. Our analysis set the decision boundaries at 68.12 pg / ml (CACTUS) and above 20.89 pg / ml (random forest), which is comparable to previously obtained data and allows a more detailed classification of patients. It is important to note that IL-8, as a pro- inflammatory cytokine, is also elevated in the samples of patients with urinary tract infections (Ko et al., 1993), so it should be used more as a biomarker of pathological conditions rather than specific diseases.

[0108] Several biomarkers important for bladder cancer screening were also indicated in the models in the male dataset, i.e., BTA (CACTUS), HAD (random forest and decision tree), and S100 calcium-binding protein A4 (S100A4) (random forest), however they were not among the most important features in the respective models. This may be due to the wide variety of diseases underlying haematuria in the datasets. BTA (Guo et al., 2014; Sharma et al., 1999), HAD (Lokeshwar et al., 2000) and S100A4 (Sagara et al., 2010) are closely related to tumourigenesis, so the addition of samples from individuals without malignant disease could influence the distribution of these features, making them less important for the classification process.

[0109] Methods

[0110] HaBio Cohort

[0111] The Haematuria Biomarker (HaBio) Study was a three-way collaborative project between Queen's University Belfast, Northern Ireland Health Trusts and Randox Laboratories Ltd. HaBio was funded by Invest Northern Ireland and Randox Laboratories Ltd. Ethical approval was obtained from the Office for Research Ethics Committee Northern Ireland (ll / NI / 0164) to recruit patients who satisfied the HaBio study inclusion criteria (Duggan et al., 2022). The protocol for HaBio was also reviewed by hospital review boards and was conducted according to the Standards for Reporting of Diagnostic Accuracy (STARD) (Bossuyt et al., 2003). A total of n=677 patients were recruited to HaBio, of which n=2 patients were excluded due to incomplete data. Therefore, the complete dataset is available for n=675 patients (n=485 males and n=190 females). There are significantly more males (2.5:1 ratio of males to females) which reflect "real world" urology patterns of presentation to haematuria clinics at the time of recruitment. This observation is borne out by the large number of men with benign prostatic hyperplasia (BPH) as a cause for haematuria. Within each gender there was a 2:1 ratio of non-cancer versus cancer (males 1.9:1 (319:166); females 2.7:1 (139:51), Figure 1).

[0112] Inclusion criteria

[0113] Bladder cancer patients -

[0114] • Written informed consent to participate in the study

[0115] • Aged between 40 and 80 years

[0116] • Current haematuria or a history of haematuria

[0117] • Cystoscopy within the last 6 months or planned cystoscopy

[0118] • No chemo- or radio- therapy in the three weeks prior to recruitment

[0119] • No previous history of cancers other than bladder cancer

[0120] • Suspicion of bladder cancer or proven bladder cancer

[0121] Control patients -

[0122] • Written informed consent to participate in the study

[0123] • No previous history of cancer

[0124] • Of the same gender, approximate age, and smoking status (where possible) to a bladder cancer patient already recruited to HABIO Current haematuria or a history of haematuria

[0125] Negative cystoscopy within the last 3 months, but at least 48h after the procedure

[0126] No chemo- or radio- therapy in the three weeks prior to recruitment

[0127] Exclusion criteria

[0128] Bladder cancer patients -

[0129] • No written informed consent to participate in the study

[0130] • Aged < 40 or > 85 years

[0131] • No history of haematuria

[0132] • No recent or planned cystoscopy

[0133] • Chemo- or radio- therapy in the three weeks prior to recruitment

[0134] • Previous history of cancer(s), other than bladder cancer

[0135] • No suspicion of bladder cancer or proven bladder cancer

[0136] Control patients -

[0137] • No written informed consent to participate in the study

[0138] • Previous history of any cancer

[0139] • Not of the same gender, approximate age and smoking status of a patient already recruited as a bladder cancer patient

[0140] • No history of haematuria

[0141] • No recent or planned cystoscopy

[0142] • Chemo- or radio- therapy in the three weeks prior to recruitment

[0143] Biomarker analysis

[0144] At the time of recruitment, a research nurse or clinician measured each patient's height, weight and blood pressure while also recording details of medical history, lifestyle / behaviours, and occupations before collecting urine (25ml) and blood (35ml) samples. In the collected samples, 80 biomarkers previously indicated as potential biomarkers of urinary tract diseases, representing a range of biological pathways (Table 1), were measured. Patient samples were analysed in triplicate and the results were expressed as a mean ± SD.

[0145] In the study, chosen biomarkers were analysed with several different techniques. At recruitment, patient urine samples were collected prior to cytoscopic examination and evaluated using the POC test for NMP22 (BladderChek, Alere, US). Osmolarity (mOsm) was determined using a Loser Micro- osmometer according to manufacturer's instructions (Loser Messtechnik, Berlin, Germany). Total urinary protein levels (mg / ml) were measured by Bradford assay (Pierce, Rockford, IL, USA). For multimarker analysis Biochip Array Technology was used (simultaneous detection of multiple analytes from a single patient urine and / or serum sample) (Randox Clinical Laboratory Services (RCLS), Antrim, Northern Ireland, UK), other biomarkers were measured using commercially available ELISA kits. Detailed description of analytical procedures is shown in the sections below. When data was below the Limit of Detection (LOD) or the Mean Detectable Dose (MDD) for any given test, 90% of the LOD or the MDD was used in lieu of the actual value for analysis (Duggan et al., 2022).

[0146] Biomarkers measured - 8-OHdG - 8-Hydroxydeoxyguanosine, ACR - Albumi Creatinine Ratio, BTA - Bladder Tumour Antigen, CD44 - Cell Surface Glycoprotein, CEA - Carcinoembryonic antigen, CK-18 - Cytokeratin 18, CK-20 - Cytokeratin-20, CRP - C-Reactive Protein, CXCL16 - Chemokine (C-X-C motif) Ligand 16, EGF - Epidermal Growth Factor, FABP-A - Fatty-Acid Binding Protein-Adipocyte, FAS - Fas Cell Surface Death Receptor, GRO-a - Chemokine Ligand 1, HDL - High Density Lipoprotein, HAD - Hyaluronic acid, IFNy - Interferon Gamma, IL - Interleukins, IL-12p70 - lnterleukin-12p70, LDL - Low Density Lipoprotein, LASPI - LIM and SH3 Domain Protein 1, M2PK - Muscle Type-2 Pyruvate Kinase, M30 - M30 Apoptosense, MCP-1 - Monocyte Chemoattractant Protein-1, MMP-9 - Matrix Metalloproteinase-9, MMP-9 / NGAL complex - Matrix Metalloproteinase-9 / Neutrophil-Associated Gelatinase Lipocalin Complex, MMP-9 / TIMP-1 - Matrix Metalloproteinase-9 / Tissue Inhibitor of Metalloproteinases-1 Complex, NGAL - Neutrophil-Associated Gelatinase Lipocalin, NMP22 - Nuclear Matrix Protein 22, NSE - Neuron Specific Enolase, pERK - Phospho-Extracellular Signal Regulated Kinase, PAI-l / tPA - Plasminogen Activator Inhibitor Type-l / Tissue Plasminogen Activator Complex, PSA / tPSA - Prostate-Specific Antigen / total Prostate-Specific Antigen, TGF-pi - Transforming Growth Factor-pi, S100A4 - S100 Calcium Binding Protein A4, slL-2Ra - Soluble lnterleukin-2 Receptor Alpha Chain (CD25), si L-6R - Soluble lnterleukin-6 Receptor (CD126), TN F-a - Tumour Necrosis Factor-Alpha, sTNFRI and II - Soluble Tumour Necrosis Factor Receptor-1 and -2, TM - Thrombomodulin, tPA - Tissue Type Plasminogen Activator, VEGF - Vascular Endothelial Growth Factor.

[0147] Analytical procedures

[0148] Patient samples were run in triplicate and the results are expressed as a mean + SD (n=3). Biochip Array Technology (Randox Clinical Laboratory Services (RCLS), Antrim, Northern Ireland, UK) was used for the simultaneous detection of multiple analytes from a single patient sample (urine or serum). Samples were analysed following manufacturer's instructions (Randox Laboratories Ltd, Crumlin, UK).

[0149] Biochip Array Technology and Biomarker Limits of Detection Biochip Array Technology (BAT) was used by Randox Clinical Laboratory Services (RCLS), Randox Science Park, Antrim, UK by scientists blinded to patient data, for the simultaneous detection of multiple biomarkers from a single patient sample1. The analytical sensitivity of the biochip(s) was as follows: cystatin C 0.60 ng / ml; EGF 2.5 pg / ml; I FNy 2.1 pg / ml; IL-24.8 pg / ml; IL-2Ra 0.12 ng / ml; IL-23 13.0 pg / ml; IL-3 8.78 pg / ml; IL-4 6.6 pg / ml; IL-6 1.2 pg / ml; IL-6R 0.62 ng / ml; IL-7 1.11 pg / ml; IL-8 7.9 pg / ml; IL-10 1.1 pg / ml; IL-12p70 2.61 pg / ml; IL-13 5.23 pg / ml; VEGF 14.6 pg / ml; TNFa 4.4 pg / ml; IL- la 0.8 pg / ml; IL-ip 1.6 pg / ml; MCP-1 13.2 pg / ml; NSE 0.26 ng / ml; NGAL 17.8 ng / ml; sTNFRl 0.24 ng / ml; D-dimer 2.1 ng / ml; sTNFR2 0.2 ng / ml; and CRP 0.67 mg / ml. Functional sensitivity for CEA and PSA (free and total) on the biochip were 0.29, 0.02 and 0.45 ng / ml, respectively. All biochips were run on an Evidence Investigator analyser according to manufacturer's instructions (Randox Laboratories Ltd, Crumlin, UK). The analytical sensitivity for HDL, LDL and cholesterol were as follows: direct HDL cholesterol (HDL) 0.189 mmol / l (7.30 mg / dl), direct LDL cholesterol (LDL) 0.189 mmol / l (7.30 mg / dl) and cholesterol 0.865 mmol / l (33.4 mg / dl), respectively. HDL, LDL and cholesterol were run on a Daytona analyser (Randox Laboratories Ltd, Crumlin, UK). The analytical sensitivity for urinary microalbumin was 5.11 mg / l. Microalbumin was analysed on a Daytona Plus analyser (Randox, Crumlin, UK). The analytical sensitivity for prolactin was 6.52 mlU / l. Prolactin was run on an Evidence Evolution analyser (Randox, Crumlin, UK). The analytical sensitivity for cystatin C was 0.4 mg / l. Serum Cystatin C was run on a Daytona analyser (RCLS, Antrim, UK). Triglycerides were run on a Daytona analyser (RCLS, Antrim, UK). Creatinine ( .mol / 1) measurements were performed by Randox Testing Services, Crumlin, UK, using a quantitative in vitro diagnostic assay from Randox (Crumlin, UK) on a Daytona analyser, according to manufacturers' instructions (Randox). The creatinine assay is linear up to 66,000 p.mol / 1 and has a lower sensitivity of 311 p.mol / 1.

[0150] Commercial ELISA kits

[0151] The following biomarkers were detected using commercially available ELISA kits, as per manufactures instructions; all patient samples were run in triplicate: 8-hydroxy 2 deoxyguanosine (8OHdG), MDD 0.1 ng / ml (Cell Biolabs, San Diego, US); Bladder tumour antigen (BTA), MDD 0.65 U / ml (Polymedco, New York, US); Cluster of differentiation 44 (CD44), MDD <0.113 ng / ml (Abeam, Cambridge, UK); UBC II (CK-8, CK-18), MDD 0.1 ng / ml (IDL, Bromma, Sweden); Cytokeratin-20, MDD 0.1 ng / ml (CK-20) (BlueGene, Shanghai, China); Clusterin, MDD 0.189 ng / ml (R&D Systems, Abingdon, UK); CXCL16, MDD 0.007 ng / ml (R&D Systems, Abingdon, UK); Cystatin B, MDD 0.013 ng / ml (R&D Systems, Abingdon, UK); Epithelial growth factor (EGF), MDD 25 pg / ml (Randox, Antrim, UK); Fatty acid-binding protein - adipose (FABP-A), MDD 0.05 ng / ml (Biovendor, Abingdon, UK); Tumour necrosis factor receptor superfamily member 6 (FAS), 5 pg / ml (RayBio, Georgia, US); Hyaluronic acid (HAD), MDD 0.1 U / l (MyBioSource, San Diego, US); C-X-C ligand 1 motif / growth regulated alpha protein (CXCLl / GROa) MDD 10 pg / ml (R&D Systems, Abingdon, UK); Interleukin 18 (IL-18), MDD 42.8 pg / ml (Randox, Antrim, UK); LIM and SH3 domain (LASP-1), MDD 6.25 pg / ml (Cusabio, Houston, US); Muscle type-2 pyruvate kinase (M2-PK), MDD 3.4 ng / ml (Randox, Antrim, UK); Caspase-cleaved CK-18 fragments (M30), MDD 20 U / L (Previva, Paudex, Switzerland); Midkine, MDD 8 pg / ml (CellMid, Sydney, Australia); Matrix metallopeptidase 9 / Neutrophil gelatinase-associated lipocalin complex (MMP-9 / NGAL), MDD 0.013 ng / ml (R&D Systems, Abingdon, UK); Matrix metallopeptidase 9 / Tissue inhibitor of metallopeptidase- 1 complex (MMP-9 / TIMP-1), MDD 0.0469 ng / ml (R&D Systems, Abingdon, UK); Plasminogen activator inhibitor-l / Tissue plasminogen activator complex (PAI-l / tPA), MDD 0.04 ngml (AssayPro, Missouri, US); Phospho-extracellular signal-related kinase (pERK), MDD 18.75 pg / ml (MyBioSource, San Diego, UK); Progranulin, MDD 0.17 ng / ml (R&D Systems, Abingdon, UK), S100 calcium-binding protein A4 (S100A4), MDD 0.225 ng / ml (Cusabio, Houston, US); Transforming growth factor beta-1 (TGFpi), MDD 4.61 pg / ml (R&D Systems, Abingdon, UK); Thrombomodulin, MDD 7.82 pg / ml (R&D Systems, Abingdon, UK) and Tissue plasminogen activator (TPA), MDD 0.01 ng / ml (Abeam, Cambridge, UK).

[0152] Point of care assays and investigations

[0153] At recruitment, patient urine samples were collected prior to cytoscopic examination and evaluated using the point of care test (POCT) Nuclear Matrix Protein 22 (NMP22) (BladderChek, Alere, US), according to manufacturer's instructions (MDD <10 U / ml were negative). Aution sticks IDEA, used for dipstick urinalysis, were interpreted using a PocketChem analyser (Arkray Inc, Japan).

[0154] Osmolality

[0155] Osmolality (mOsm) was determined using a Loser Micro-osmometer according to manufacturer's instructions (Loser Messtechnik, Berlin, Germany).

[0156] Total Urinary Protein (Bradford Assay)

[0157] Total urinary protein levels (mg / ml) were determined, in triplicate, by Bradford assay (Pierce, Rockford, IL, USA) using a stock solution of BSA (Sigma) as standard (1 mg / ml). Patient urine samples (10 pl / patient), after centrifugation (1200 g, 10 minutes, 4°C), were mixed with Bradford reagent (1 ml) and allowed to stand for 5 minutes. The samples were read on a Hitachi Spectrophotometer (Model No. U-2800) at Asesnm. Total urinary protein was determined using a BSA calibration chart.

[0158] Data below the Limit of Detection (LOD) / Mean Detectable Dose (MDD) - when data was below the LOD / MDD for any given test, 90% of the LOD / MDD was used in the analysis. Table 1. Biological pathways associated with measured biomarkers.

[0159] Classical approach

[0160] In the study, due to differences in the typical causes of haematuria and the prevalence of malignant diseases we analysed data separately for male and female participants, in addition to the entire cohort. In the pre-processing step, as the data were characterised by a highly skewed distribution, we performed a log transformation of the biomarker measurement results for further analysis to reduce the skewness and replaced missing data with the median value for the given biomarker. For analysis, urine and serum biomarkers were used; if the same biomarker was analysed in both serum and urine samples, serum results are indicated by the word "serum" in the biomarker name.

[0161] Firstly, we performed k-means clustering to assess if analysed features could be linearly separated. For k-means clustering, we iteratively tested the number of clusters from 1 to 20 and used the silhouette width to select the best configuration. We observed that for all three data subsets, the optimal value of clusters for k-means clustering was 2, showing that the distribution of features does not follow clear macro patterns or reflect the underlying number of causes of haematuria (Figure 2). As it was not possible to distinguish the number of clusters reflecting the number of underlying classes of final diagnosis, we decided to stratify patients into two subgroups, sick and healthy. The sick population had any of the following possible causes for their haematuria: chronic kidney diseases, infection, other benign diagnosis, bladder cancer, history of bladder cancer or other types of cancer (e.g., prostate cancer, renal cell carcinoma). The healthy population included every patient with no causes identified for their haematuria.

[0162] Initial analysis involved logistic regression analysis and assessment of balanced accuracy for each biomarker separately. Afterwards, we applied binary decision trees and random forest models. For both models we performed 10-fold cross validation repeated three times. As the random forest is not an inherently explainable method, in contrast to the decision trees, we applied the local interpretable model-agnostic explanations (LIME) algorithm (Ribeiro et al., 2016), to provide the explanation of the classification process and understanding of the biomarkers' influence on the final class prediction. LIME focuses on explaining the model's prediction for individual cases. LIME generates a new dataset consisting of perturbed samples and the corresponding predictions and then trains an interpretable model (regression) on this new dataset, weighted by the proximity of the sampled cases to the case of interest. Because a linear model is inherently interpretable, the fitted weights can be inspected and viewed as proxies for feature importance and based on the proximity of the values to the perturbed data point, the cut-off values for individual features can be provided.

[0163] All the analysis was performed in R (R Core Team, 2023).

[0164] CACTUS classification

[0165] To model the healthy and sick classes, we used the CACTUS (Gherardini et al., 2024) algorithm. In the first step, fully anonymized data abstractions of the quantitative and qualitative biomarker data were generated by (Barabasi et al., 2011) transforming raw biomarker data into two-stage data abstractions (flips) based on receiver-operator curve (ROC) theory. These flips were encoded with the last letter of the label for each biomarker: up (U) abstracts raw data above, and down (D) below calculated cut-off values. For each biomarker, significance was determined from the node's conditional probability P(f \ ci) of the flip f, given the class ci (sick or healthy). To assess how the conditional probability P(f | ci), will change across the N considered classes, and to infer their importance for the classification process, ranks (Rxf) were calculated for each biomarker according to the following equation:

[0166] To assess the accuracy of the network's patient classification we calculated the cost function. For every patient in the given state "s" (sick or healthy) the cost function (Cs) was calculated based on the corresponding node significance (<JS,I) of each biomarker (xi):

[0167] The cost function with the greater value was the determinant for patient classification as sick or healthy. Obtained classifications were compared to real diagnosis groups, marked as true positive (TP), false negative (FN), true negative (TN) or false negative (FN) and used to calculate specificity (Eq.3. a), sensitivity (Eq.3.b) and accuracy (Eq.3.c) for all tested models. Due to the much higher number of sick patients in our study groups, we used balanced accuracy (a metric which is robust for unbalanced datasets) to assess model performance.

[0168] „ . . .

[0169] Sensitivity (Eq.3.a)

[0170] Specificity (Eq.s.b)

[0171] . senstivity+specificity

[0172] Balanced accuracy = - y2- — - — - (Eq.s.c)

[0173] Results

[0174] The results for the stratification of haematuria patients into sick and healthy categories using different statistical methods are presented below:

[0175] Assessment of the balanced accuracy of the logistic regression showed that single biomarkers were not specific enough to discriminate between healthy and sick patients (Tables 2 and 3). For both genders, the highest accuracy was obtained for soluble tumour necrosis factor receptor I (sTNFRI, 0.572), cystatin C (0.569) and progranulin (0.516). For females (Table 2), the three biomarkers with the best scoring performance (based on balanced accuracy) were phospho-extracellular signal regulated kinase (pERK, 0.673), microalbumin (0.672) and chemokine (C-X-C motif) ligand 16 (CXCL16, 0.667). In the male data subset (Table 3), which is highly unbalanced, no single biomarker gave an accuracy higher than 0.5. Logistic regression results for combinations of the four preferred biomarkers of the invention are shown in Table 4.

[0176] Table 2. Logistic regression classification results for every marker in different types of samples (urine or serum) for females. Number of cases in the sick class: 111, number of cases in healthy class: 79. Sick class was assumed to be positive.

[0177] Table 3. Logistic regression classification results for every marker in different types of samples (urine or serum) for males. Number of cases in the sick class: 441, number of cases in healthy class: 44. Sick class was assumed to be positive.

[0178] Table 4. Logistic regression classification results for the four preferred urine markers and combinations in female haematuria patients. Number of cases in the sick class: 111, number of cases in healthy class: 79.

[0179] Decision trees provided simple rule-based models based on a maximum of 14 biomarkers (including gender) for patient classification. The most complicated tree was built for a dataset with patients of both genders. The first branch was built based on the male gender; resulting in the subsequent branches being gender specific. The male and female decision trees were similar to the branches of the tree built for both genders, with some additional branches. In the case of males, stratification was improved by adding decision boundaries based on serum hyaluronic acid (HAD) and pERK levels, allowing additional healthy individuals to be distinguished. In the female decision trees, the situation was reversed; the classification was performed with a lower number of features and some of the branches of the trees for both genders, such as vascular endothelial growth factor (VEGF), were pruned. The highest balanced accuracy of decision tree classification was obtained when both genders were analysed together (0.640), even though the first split was on gender. Separate stratification for males and females gave lower balanced accuracy (0.551, and 0.623), and better significance and specificity was obtained for females as the data subset was more balanced (Table 5).

[0180] Table 5. Decision tree classification results for the four preferred urine markers and combinations in female haematuria patients. Number of cases in the sick class: 111, number of cases in healthy class: 79.

[0181] The effectiveness of the random forest classification was also insufficient to discriminate between sick and healthy individuals. The highest value of balanced accuracy was obtained for the female data subset (0.665), lower for both genders (0.627) and the lowest for the male subset (0.512). The corresponding values of sensitivity and specificity showed a bias towards the more prevalent class (sick), which is most visible in the case of the male data subset (sensitivity: 1.00, specificity: 0.024), showing that all cases of sick individuals were correctly classified and only one healthy individual was correctly classified. We also extracted the top 10 features for random forest classification, two of the most important features are gender-specific (serum total prostate specific antigen (tPSA) and gender), justifying the need to separate female and male cases for analysis purposes. Several biomarkers such as: microalbumin, osmolarity, sTNFRI, cystatin C, CXCL16, pERK, progranulin, and patient age were of high importance for two or three data subsets. Although the biomarkers were common to the data subsets, as the LIME analysis showed, the decision boundaries (levels of the biomarkers) and their contribution (weights) to the final model were different. For example, for age, which is one of the most important characteristics, the lowest cut-off value in the case of both genders was 60 and has a slightly higher influence on the classification of the healthy category than the sick category; for the male data subset only, this cut-off value was 61, with the same influence on the classification. In the case of all analysed data subsets, there were some biomarkers within certain ranges that had a clear positive or negative influence on the classification. CACTUS classification gave a higher balanced accuracy than the models described above for all analysed data subsets (0.747 both genders, 0.83 females, 0.718 males). Moreover, the obtained values of sensitivity and specificity were more balanced, although the sensitivity was lower, which indicated a higher false negative rate. The CACTUS specificity was higher than the specificity for decision trees and random forests, showing that the classification was not biased towards the predominant group (sick individuals) (Table 6).

[0182] Table 6. CACTUS classification results for the four preferred urine markers and combinations in female haematuria patients. Number of cases in the sick class: 111, number of cases in healthy class: 79.

[0183] The 10 biomarkers with the highest CACTUS ranks for sick and healthy individuals in females is shown in Figure 2. The ranks provide information about the average difference between the classes (sick and healthy) for the probability of the biomarkers being in each state ('U' or 'D'), meaning that the higher the rank value, the greater the difference in at least one of the probabilities. Like random forest, CACTUS confirmed the need to stratify patients into subgroups based on gender, as gender was indicated by CACTUS as the most important factor for the whole population studied. Additionally, the second most important biomarker was serum tPSA, a gender-specific biomarker of prostate health and therefore important in the classification process. Microalbumin was reported as the third most important biomarker for both genders, but also received a high score in the gender stratified analysis (second for men and first for women).

[0184] In the male population, age is the highest ranked factor and was not reported in any of the other subsets analysed. In addition, as in the random forest analysis, several biomarkers such as serum tPSA, microalbumin, CXCL16, urinary protein, monocyte chemoattractant protein-1 (MCP-1), and progranulin were present with a high score in more than two groups and are therefore more sensitive to discovering the differences in flips which were more prevalent in the healthy class, than the sick class. This was visible as a higher difference in between probabilities of nodes in the healthy. Interestingly, microalbumin was the only common biomarker in both the male and female results, suggesting a gender-specific mechanism for haematuria development. The flip probabilities indicated whether the biomarker was generally below ('D') or above ('U') the calculated cut-off values; we observed that the distribution of some of the biomarkers changed significantly between classes. For example, when looking at the dataset with both genders, we can see that being male, having a serum tPSA above the cut-off value and having high levels of microalbumin are important factors for classification as sick. We have also observed that some features were only important for classification in one class and for the second class there was an equal or almost equal probability of the flip probabilities. This was the case in the both genders dataset for sTNFRI, which had the same probability of flip for healthy individuals (0.5, 0.5), but showed higher probability (0.836) for sTNFRI being in state ('U') for the patients classified as sick. In the male population, the probability of flips indicated that being older (0.721), with higher levels of MCP-l (0.797), serum tPSA (0.649) and sTNFRII (0.836) and reduced levels of D-dimer (0.676) were important factors in being sick. It is interesting to note that in the male population, CACTUS classification detected higher differences in the flip's probability for healthy individuals than in sick. In the female dataset we observed the gradual change in the flip probabilities, i.e., the highest difference in the flip probability, which was most important for healthy individuals (microalbumin), had at the same time the lowest difference for sick individuals and vice versa (Figure 3).

Claims

Claims1. A method for aiding in the diagnosis of a female patient presenting with haematuria, said method comprising (i) determining in a sample previously isolated from said patient, the concentration of one or more biomarkers selected from the list consisting of Phospho- Extracellular Signal Regulated Kinase (pERK), Chemokine (C-X-C motif) Ligand 16 (CXCL16), microalbumin and interleukin 8 (IL-8); (ii) stratifying the patient into "healthy" or "sick" categories wherein the detection of an elevated concentration of the one or more biomarkers compared to a normal control indicates that the patient should be placed into the "sick" category.

2. The method of claim 1 wherein the panel of biomarkers comprises pERK.

3. The method of claim 1 wherein the panel of biomarkers comprises CXCL16.

4. The method of claim 1 wherein the panel of biomarkers comprises Microalbumin.

5. The method of claim 1 wherein the panel of biomarkers comprises IL-8.

6. The method of claim 1 wherein the panel of biomarkers is selected from: (i) pERK and CXCL16; (ii) pERK, CXCL16 and IL-8; (iii) pERK, CXCL16 and Microalbumin; or (iv) pERK, CXCL16, microalbumin and IL-8.

7. The method of any preceding claim wherein the concentration of one or more additional biomarkers is determined, said biomarkers selected from the list consisting of EGF, VEGF, I Lla, Creatinine, Cystatin C, sTNFRl, progranulin, ACR and midkine.

8. The method of any preceding claim, wherein one or more of the biomarkers are detected in a urine sample.

9. The method of claim 8 wherein each biomarker is detected in a urine sample.

10. The method of any preceding claim wherein step (ii) comprises inputting the measured concentrations of the biomarkers from step (i) into a statistical algorithm wherein the output of the algorithm indicates whether the patient should be placed into the "healthy" or "sick" categories.

11. A solid support material comprising binding probes attached thereto, said binding probes having specific affinity for one or more biomarkers selected from pERK, CXCL16, microalbumin and IL-8.

12. The solid support material of claim 11 comprising binding probes attached thereto, said binding probes having specific affinity to each biomarker in a panel of biomarkers selected from: (i) pERK and CXCL16; (ii) pERK, CXCL16 and IL-8; (iii) pERK, CXCL16 and Microalbumin; or (iv) pERK, CXCL16, microalbumin and IL-8.

13. The solid support material of claims 11 - 12 wherein the binding probes are antibodies or fragments thereof.

14. The solid support material of claims 11 - 13 wherein the support is a biochip.

15. A method to aid in the diagnosis of bladder cancer in a female patient presenting with haematuria, said method comprising detecting the concentration of one or more biomarkers selected from the list comprising pERK, CXCL16, microalbumin and IL-8 in a sample previously isolated from said patient; wherein an elevated level of one or more of the biomarkers compared to a normal control is indicative of risk of bladder cancer.

16. The method of any previous claim comprising an additional step of determining if the patient has an infection.

17. A method for aiding in the diagnosis of a male patient presenting with haematuria, said method comprising (i) determining in a sample previously isolated from said patient, the concentration of one or more biomarkers selected from the list consisting of HAD, microalbumin, tPSA, cystatin C, BTA and S100A4; (ii) stratifying the patient into "healthy" or "sick" categories wherein the detection of an elevated concentration of the one or more biomarkers compared to a normal control indicates that the patient should be placed into the "sick" category.

Images