Jonathan Shoag

None

Resident Physician

PLCO (Learn more about this study)

PLCO-90

May 28, 2014

Use of Regression Discontinuity to Measure the Effect of PSA Screening

One of the criticisms of the PLCO prostate cancer screening trial is the rate of PSA screening in the control arm. Regression discontinuity (RD), a statistical technique that has been used increasingly in economics, utilizes the discrete nature of treatment to capture “random” variation, and thus may provide a method to reanalyze PLCO data which avoids this issue. RD utilizes the idea that a continuous endogenous variable along which treatment is assigned generally demonstrates continuous effects. Because treatment is discrete, examining the region around an assignment cutoff allows the determination of treatment effect. The small variation in the continuous variable should not have much impact, so the observed effect should be the effect of the treatment.

We would like to look at the effect of screening in the screening arm only by using the discrete cutoff, of a PSA of 4.0 ng per milliliter (at which individuals were told they had prostate cancer and follow up was recommended) in order to assess the effect of screening (a simplified example is if individuals in the screening arm with a PSA of 3.9 had a higher CaP specific mortality than those with a PSA of 4.1 it would suggest that screening was beneficial). This will allow us to instrument for treatment using this discrete cutoff, and thereby introduce a new "randomization" for treatment. By examining if there is a discontinuity we should be able to determine if there is a benefit to screening.

1. Estimate the efficacy of PSA screening using a regression discontinuity approach.

2. Introduce regression discontinuity as a statistical technique in clinical research