Use of percent free PSA (%fPSA) and artificial neural networks (ANNs) can eliminate unnecessary prostate biopsies. In a total of 4,480 patients from five centers with PSA concentrations in the range of 2-10 ng/ml an IMMULITE PSA-based ANN (iANN) was compared with other PSA assay-adapted ANNs (nANNs) to investigate the impact of different PSA assays. ANN data were generated with PSA, fPSA (assays from Abbott, Beckman, DPC, Roche or Wallac), age, prostate volume, and DRE status. In 15 different ROC analyses, the area under the curve (AUC) in the PSA ranges 2-4, 2-10, and 4-10 ng/ml for the nANN was always significantly larger than the AUC for %fPSA or PSA. The nANN and logistic regression models mostly also performed better than the iANN. Therefore, for each patient population, PSA assay-specific ANNs should be used to optimize the ANN outcome in order to reduce the number of unnecessary biopsies.
Use of percent free PSA (%fPSA) and artificial neural networks (ANNs) can eliminate unnecessary prostate biopsies. In a total of 4,480 patients from five centers with PSA concentrations in the range of 2-10 ng/ml an IMMULITE PSA-based ANN (iANN) was compared with other PSA assay-adapted ANNs (nANNs) to investigate the impact of different PSA assays. ANN data were generated with PSA, fPSA (assays from Abbott, Beckman, DPC, Roche or Wallac), age, prostate volume, and DRE status. In 15 different ROC analyses, the area under the curve (AUC) in the PSA ranges 2-4, 2-10, and 4-10 ng/ml for the nANN was always significantly larger than the AUC for %fPSA or PSA. The nANN and logistic regression models mostly also performed better than the iANN. Therefore, for each patient population, PSA assay-specific ANNs should be used to optimize the ANN outcome in order to reduce the number of unnecessary biopsies.