BACKGROUND: Diagnosis of prostate cancer at an early stage can potentially identify tumours when intervention may improve treatment options and survival. AIM: To develop and validate an equation to predict absolute risk of prostate cancer in asymptomatic men with prostate specific antigen (PSA) tests in primary care. DESIGN AND SETTING: Cohort study using data from English general practices, held in the QResearch database. METHOD: Routine data were collected from 1098 QResearch English general practices linked to mortality, hospital, and cancer records for model development. Two separate sets of practices were used for validation. In total, there were 844 455 men aged 25-84 years with PSA tests recorded who were free of prostate cancer at baseline in the derivation cohort; the validation cohorts comprised 292 084 and 316 583 men. The primary outcome was incident prostate cancer. Cox proportional hazards models were used to derive 10-year risk equations. Measures of performance were determined in both validation cohorts. RESULTS: There were 40 821 incident cases of prostate cancer in the derivation cohort. The risk equation included PSA level, age, deprivation, ethnicity, smoking status, serious mental illness, diabetes, BMI, and family history of prostate cancer. The risk equation explained 70.4% (95% CI = 69.2 to 71.6) of the variation in time to diagnosis of prostate cancer (R2) (D statistic 3.15, 95% CI = 3.06 to 3.25; Harrell's C-index 0.917, 95% CI = 0.915 to 0.919). Two-step approach had higher sensitivity than a fixed PSA threshold at identifying prostate cancer cases (identifying 68.2% versus 43.9% of cases), high-grade cancers (49.2% versus 40.3%), and deaths (67.0% versus 31.5%). CONCLUSION: The risk equation provided valid measures of absolute risk and had higher sensitivity for incident prostate cancer, high-grade cancers, and prostate cancer mortality than a simple approach based on age and PSA threshold.
Br J Gen Pract
e364 - e371
cohort studies, primary health care, prostate cancer, prostate-specific antigen, risk prediction