A System Dynamics Model of Serum Prostate-Specific Antigen Screening for Prostate Cancer

Abstract

Since 2012, US guidelines have recommended against prostate-specific antigen (PSA) screening for prostate cancer. However, evidence of screening benefit from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening trial and the European Randomized Study of Screening for Prostate Cancer has been inconsistent, due partly to differences in noncompliance and contamination. Using system dynamics modeling, we replicated the PLCO trial and extrapolated follow-up to 20 years. We then simulated 3 scenarios correcting for contamination in the PLCO control arm using Surveillance, Epidemiology, and End Results (SEER) incidence and survival data collected prior to the PSA screening era (scenario 1), SEER data collected during the PLCO trial period (1993-2001) (scenario 2), and data from the European trial's control arm (1991-2005) (scenario 3). In all scenarios, noncompliance was corrected using incidence and survival rates for men with screen-detected cancer in the PLCO screening arm. Scenarios 1 and 3 showed a benefit of PSA screening, with relative risks of 0.62 (95% confidence interval: 0.53, 0.72) and 0.70 (95% confidence interval: 0.59, 0.83) for cancer-specific mortality after 20 years, respectively. In scenario 2, however, there was no benefit of screening. This simulation showed that after correcting for noncompliance and contamination, there is potential benefit of PSA screening in reducing prostate cancer mortality. It also demonstrates the utility of system dynamics modeling for synthesizing epidemiologic evidence to inform public policy.

Keywords: cancer-specific mortality; policy evaluation; prostate cancer; prostate-specific antigen screening; system dynamics modeling.

Figure 1.

Stock-and-flow diagram for a system dynamics model of prostate cancer (PrCa) screening in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial. Stocks, denoted by boxes, represent accumulations of subjects. Flows, denoted by pipes with valves, represent rates of movement between stocks. Auxiliaries, denoted by an italic font, represent variables in equations used to calculate rates. For simplicity, arrows linking stocks that are used to calculate rates were omitted from the diagram. Cancer stage was defined using the Surveillance, Epidemiology, and End Results categories as follows: localized (stages I and II), regional (stage III), distant (stage IV), and unstaged/unknown. CA, control arm; SA, screening arm.

Figure 2.

Simulated numbers of incident prostate cancer cases under different scenarios in the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. A) Base case scenario: replication of results from the PLCO trial. B) Scenario 1: results corrected to Surveillance, Epidemiology, and End Results (SEER) data from the era before prostate-specific antigen screening (1975–1988). C) Scenario 2: results corrected to SEER data from the period 1993–2001. D) Scenario 3: results corrected to data from the European Randomized Study of Screening for Prostate Cancer control arm (1991–2005). Areas shaded in gray represent data that were extrapolated (from 13 years of follow-up to 20 years of follow-up).

Figure 3.

Simulated numbers of prostate cancer deaths under different scenarios in the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. A) Base case scenario: replication of results from the PLCO trial. B) Scenario 1: results corrected to Surveillance, Epidemiology, and End Results (SEER) data from the era before prostate-specific antigen screening (1975–1988). C) Scenario 2: results corrected to SEER data from the period 1993–2001. D) Scenario 3: results corrected to data from the European Randomized Study of Screening for Prostate Cancer control arm (1991–2005). Areas shaded in gray represent data that have been extrapolated (from 13 years of follow-up to 20 years of follow-up).