Urine Biomarker for Bladder Cancer: Predictive Values and Spectrum Bias
A prospective single-center study evaluates a urine biomarker for bladder cancer in 1,000 patients with hematuria. Prevalence is 12%. Sensitivity is 88% and specificity 82% against cystoscopy plus biopsy. The authors conclude the test is 'highly accurate and suitable for ruling out cancer.' A second validation set is a case-control study of 500 known bladder cancer cases and 500 healthy controls, which reports sensitivity 94% and specificity 91%. The hospital laboratory committee asks whether to offer the test and how to interpret the two studies. The assay is CLIA-validated in-house and costs $180; cystoscopy costs $600. The first study had 20% low-grade non-muscle-invasive tumors; 8% were lost to follow-up.
1. Calculate the positive and negative predictive values in the prospective study and interpret them for clinical use.
Read the worked answer
At 12% prevalence, PPV = (0.88 × 0.12) / [(0.88 × 0.12) + (0.18 × 0.88)] = 0.1056 / 0.264 = 40%. NPV = (0.82 × 0.88) / [(0.82 × 0.88) + (0.12 × 0.12)] = 0.7216 / 0.736 = 98%. So a positive result is wrong in about 60% of patients, while a negative result is reassuring but not perfect.
The 'rule out' claim needs qualification. An NPV of 98% means 2% of test-negative patients still have cancer; in a 1,000-patient hematuria population that is about two missed cancers. The test could reduce cystoscopy in low-risk patients, but it cannot replace cystoscopy in high-risk patients or those with ongoing symptoms. The PPV of 40% means the test is poor as a stand-alone diagnostic test; it would generate many unnecessary cystoscopies if used that way.
2. How do the design features of the two studies affect the accuracy estimates?
Read the worked answer
The prospective study is stronger because it enrolled consecutive symptomatic patients, but it still has limitations: 8% loss to follow-up can bias estimates, and the 20% low-grade tumors may make sensitivity look better or worse depending on assay performance. The case-control study is the larger threat to validity. Cases are known cancer patients and controls are healthy, so the spectrum is unrepresentative; this spectrum bias typically inflates both sensitivity and specificity. It cannot provide a valid PPV because the artificial 50% prevalence is not the clinical prevalence.
I would not average the two studies or choose the more favorable estimate. The case-control results should be used only as a signal, not as the basis for adoption. A prospective cohort with consecutive hematuria patients, complete reference-standard verification, and blinded readers is needed. Confidence intervals should be reported, and the study should include the clinically relevant low-grade and high-grade subgroups.
3. Would you recommend offering the test? What additional evidence and safeguards would you require?
Read the worked answer
I would not offer the test as a replacement for cystoscopy. I would consider it only as a triage adjunct in a defined low-risk pathway, with a protocol that mandates cystoscopy for positive results and for high-risk patients regardless of result. The cost difference ($180 versus $600) is attractive, but a 40% PPV means many negative cystoscopies; the health-economic case depends on the local prevalence, the cost of missed cancer, and the downstream workup.
Before adoption, I would require a prospective local validation or a decision-curve analysis showing net benefit, a clear cutoff, and a plan for incidental findings. The laboratory should monitor real-world PPV, NPV, and interval cancers. The committee should also address the 8% loss to follow-up and the case-control bias explicitly in its minutes, because adopting the 94%/91% figures would overstate performance.