Developing a Diagnostic Multivariable Prediction Model for Urinary Tract Cancer in Patients Referred with Haematuria: Results from the IDENTIFY Collaborative Study

Sinan Khadhouri*, Kevin M. Gallagher, Kenneth R. MacKenzie, Taimur T. Shah, Chuanyu Gao, Sacha Moore, Eleanor F. Zimmermann, Eric Edison, Matthew Jefferies, Arjun Nambiar, Thineskrishna Anbarasan, Miles P. Mannas, Taeweon Lee, Giancarlo Marra, Juan Gómez Rivas, Gautier Marcq, Mark A. Assmus, Taha Uçar, Francesco Claps, Matteo BoltriGiuseppe La Montagna, Tara Burnhope, Nkwam Nkwam, Tomas Austin, Nicholas E. Boxall, Alison P. Downey, Troy A. Sukhu, Marta Antón-Juanilla, Sonpreet Rai, Yew Fung Chin, Madeline Moore, Tamsin Drake, James S.A. Green, Beatriz Goulao, Graeme MacLennan, Matthew Nielsen, John S. McGrath, Veeru Kasivisvanathan, IDENTIFY Study Group

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract / Description of output

Background: Patient factors associated with urinary tract cancer can be used to risk stratify patients referred with haematuria, prioritising those with a higher risk of cancer for prompt investigation. Objective: To develop a prediction model for urinary tract cancer in patients referred with haematuria. Design, setting, and participants: A prospective observational study was conducted in 10 282 patients from 110 hospitals across 26 countries, aged ≥16 yr and referred to secondary care with haematuria. Patients with a known or previous urological malignancy were excluded. Outcome measurements and statistical analysis: The primary outcomes were the presence or absence of urinary tract cancer (bladder cancer, upper tract urothelial cancer [UTUC], and renal cancer). Mixed-effect multivariable logistic regression was performed with site and country as random effects and clinically important patient-level candidate predictors, chosen a priori, as fixed effects. Predictors were selected primarily using clinical reasoning, in addition to backward stepwise selection. Calibration and discrimination were calculated, and bootstrap validation was performed to calculate optimism. Results and limitations: The unadjusted prevalence was 17.2% (n = 1763) for bladder cancer, 1.20% (n = 123) for UTUC, and 1.00% (n = 103) for renal cancer. The final model included predictors of increased risk (visible haematuria, age, smoking history, male sex, and family history) and reduced risk (previous haematuria investigations, urinary tract infection, dysuria/suprapubic pain, anticoagulation, catheter use, and previous pelvic radiotherapy). The area under the receiver operating characteristic curve of the final model was 0.86 (95% confidence interval 0.85–0.87). The model is limited to patients without previous urological malignancy. Conclusions: This cancer prediction model is the first to consider established and novel urinary tract cancer diagnostic markers. It can be used in secondary care for risk stratifying patients and aid the clinician's decision-making process in prioritising patients for investigation. Patient summary: We have developed a tool that uses a person's characteristics to determine the risk of cancer if that person develops blood in the urine (haematuria). This can be used to help prioritise patients for further investigation.

Original languageEnglish
JournalEuropean Urology Focus
Early online date24 Jun 2022
Publication statusE-pub ahead of print - 24 Jun 2022

Keywords / Materials (for Non-textual outputs)

  • Bladder cancer
  • Haematuria
  • Prostate cancer
  • Renal cancer
  • Risk Calculator
  • Risk factors
  • Urinary tract cancer
  • Urothelial cancer


Dive into the research topics of 'Developing a Diagnostic Multivariable Prediction Model for Urinary Tract Cancer in Patients Referred with Haematuria: Results from the IDENTIFY Collaborative Study'. Together they form a unique fingerprint.

Cite this