MODELING AND POLICY DEVELOPMENT: PSA SCREENING FOR PROSTATE CANCER

Objective: The objective of this work is to bring mathematical models of prostate cancer progression and outcomes to the policymaking table. Continuous biomarkers such as PSA yield a plethora of possible test-positive criteria. Coherent comparison of the many alternatives requires a model that links biomarker growth with disease progression. Our objective in the present study is to calibrate such a model to US population incidence data and to use it to project disease outcomes for the purpose of guideline development in prostate cancer.

Methods: The model consists of two components, PSA growth and disease progression. PSA growth is estimated via mixed effects modeling of data from the Prostate Cancer Prevention Trial.  Disease progression is governed by hazard functions that depend on PSA and determine onset, metastatic spread, and clinical detection. Given PSA growth curves, progression hazards are estimated by maximum likelihood using data on prostate cancer incidence in the US from 1975 to 2000. This procedure also utilizes information on the frequency of PSA testing and biopsy following a test by age and year. We use the calibrated model to project early detections and overdiagnoses for PSA cutoffs of 2.5 versus 4ng/ml, representing an ongoing quandary in the prostate screening community.

Results: The calibrated model matches the peak in observed early stage incidence but the modeled decline in advanced disease is less pronounced than that observed.  Disease natural history measures validate well against other US population studies. The estimated mean lead times range from 4 to 7 years depending on age. Sojourn times range from 2 to 13 years. Assuming that biopsy referral for PSA between 2.5 and 4.0 ng/ml would approximate that for PSA between 4.0 and 7.0 ng/ml, we find that the lower cutoff shifts 1 case from advanced to localized stage per 422 additional overdiagnoses. Conclusions: Using a model that links biomarker growth with disease progression allows us to quantify the likely benefits associated with different screen-positive criteria and to consider the balance of benefit and cost. We conclude that a lower PSA cutoff produces little benefit in terms of detection of curable disease relative to the standard cutoff of 4 ng/ml at considerable cost.