The Stockholm-3 Model for Prostate Cancer Detection: Algorithm Update, Biomarker Contribution, and Reflex Test Potential.

Abstract

It has been shown that the Stockholm-3 model (S3M) outperforms prostate-specific antigen (PSA) as a screening tool for prostate cancer. To update the S3M, to give a detailed account of the value of each predictor in the S3M, and to evaluate the S3M as a reflex test for men with PSA ≥3ng/ml. During 2012-2015, the Stockholm-3 study evaluated the S3M relative to PSA as tests for Gleason score ≥7 prostate cancers among men aged 50-69 yr. The participants (n=59 159) underwent both tests, and biopsy was recommended if at least one was positive. A total of 5073 men had a biopsy because of elevated PSA (≥3ng/ml). Logistic regression was used to update the S3M: intact PSA was removed, HOXB13 was included, and the model was fitted to data from the Stockholm-3 training and validation cohorts. To compare S3M with PSA, we fixed the sensitivity for detection of high-grade cancer and evaluated the performance as the number of biopsies needed to achieve that sensitivity for each test. The updated S3M slightly improved the area under the receiver operating characteristic curve compared to previously published results (0.75 vs 0.74). When used as a reflex test for men with PSA ≥3ng/ml, S3M reduced the number of biopsies needed by 34% compared to the use of PSA alone, with equal sensitivity. A limitation is the ethnically homogeneous population. A major problem with PSA screening-too many unnecessary biopsies-can be mitigated if S3M is used as a reflex test. To find aggressive prostate cancer with the minimum number of negative biopsies and detection of clinically insignificant cancers, we evaluated the use of a personalized diagnostic prediction model as a second test for men with a positive prostate-specific antigen (PSA) test. We found that this two-step approach could reduce prostate biopsies by a third compared to using PSA alone.