Efficacy of Prostate-Specific Antigen Screening: Use of Regression Discontinuity in the PLCO Cancer Screening Trial | Cancer Screening, Prevention, Control | JAMA Oncology | JAMA Network
[Skip to Navigation]
Sign In
Figure.  Regression Discontinuity Analyses
Regression Discontinuity Analyses

Prostate cancer risk categories defined by D’Amico classification without prostate-specific antigen level. Graphs truncated at a maximum prostate-specific antigen of 15 ng/mL for ease of presentation (includes 99% of prostate-specific antigen levels).

Table.  Discontinuity in Selected Outcomes at PSA of 4.0 ng/mL
Discontinuity in Selected Outcomes at PSA of 4.0 ng/mL
1.
Andriole  GL, Crawford  ED, Grubb  RL  III,  et al; PLCO Project Team.  Mortality results from a randomized prostate-cancer screening trial.  N Engl J Med. 2009;360(13):1310-1319.PubMedGoogle ScholarCrossref
2.
Andriole  GL, Crawford  ED, Grubb  RL  III,  et al; PLCO Project Team.  Prostate cancer screening in the randomized Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial: mortality results after 13 years of follow-up.  J Natl Cancer Inst. 2012;104(2):125-132.PubMedGoogle ScholarCrossref
3.
Moscoe  E, Bor  J, Bärnighausen  T.  Regression discontinuity designs are underutilized in medicine, epidemiology, and public health: a review of current and best practice.  J Clin Epidemiol. 2015;68(2):122-133.PubMedGoogle ScholarCrossref
4.
Imbens  GW, Lemieux  T.  Regression discontinuity designs: a guide to practice.  J Econom. 2008;142(2):615-635.Google ScholarCrossref
5.
Imbens  K, Kalyanaraman  K.  Optimal Bandwidth Choice for the Regression Discontinuity Estimator.http://www.nber.org/papers/w14726. 2009. Accessed July 14, 2015.
6.
Nichols  A.  RD: Stata module for regression discontinuity estimation. http://ideas.repec.org/c/boc/bocode/s456888.html. 2011. Accessed July 7, 2015.
Research Letter
October 2015

Efficacy of Prostate-Specific Antigen Screening: Use of Regression Discontinuity in the PLCO Cancer Screening Trial

Author Affiliations
  • 1Department of Urology, New York Presbyterian Hospital, Weill Cornell Medical College, New York
  • 2Department of Urology, Massachusetts General Hospital, Boston
  • 3Department of Public Policy, Harvard Kennedy School, Cambridge, Massachusetts
JAMA Oncol. 2015;1(7):984-986. doi:10.1001/jamaoncol.2015.2993

The Prostate Lung Colorectal and Ovarian (PLCO) cancer screening trial randomized 76 693 men from 1993 to 2001 to usual care or annual prostate-specific antigen (PSA) screening for 6 years and annual digital rectal examination for 4 years. This study found that PSA screening results in increased detection of prostate cancer but does not reduce prostate cancer–specific or overall mortality. The findings of the PLCO cancer screening trial are controversial largely because of a high rate of PSA screening in the control group, which reached 52% by the sixth year of the trial.1,2 Despite this shortcoming, the PLCO trial is likely to remain the only major trial of PSA screening in the United States.

We used regression discontinuity (RD), a statistical technique used in the social sciences but rarely applied to clinical data, to address the above criticism.3 This technique allows us to examine the effect of PSA screening on outcomes using only the screening arm of the PLCO trial.

Methods

The statistical basis of RD has been described previously.4 Regression discontinuity allows us to leverage that a PSA of 4.0 ng/mL was used as the threshold for further workup in the PLCO trial (to convert PSA to micrograms per liter, multiply by 1). In the absence of a treatment effect, the regression of PSA and a given outcome should be continuous around the PSA cutoff. However, if a biopsy based on PSA screening affects an outcome, we would expect to find a discontinuity in the regression around a PSA of 4.0 ng/mL. Since confounders should be evenly distributed right below and above this cutoff, RD allows us to isolate the effect of screening on outcomes.

We obtained the 13-year screening and outcome data from the PLCO trial. The control arm of the study was dropped from all analyses. We used a first-degree local polynomial approach with the Imbens and Kalyanaraman mean squared error minimizing bandwidth.5 Our results are not sensitive to this bandwidth choice. We used STATA/ICv13.1 (StataCorp) for statistical analysis. An RD analysis code was generated, and we confirmed its accuracy using a Stata module for RD estimation.6 A waiver was obtained from the Office of Research Integrity at Weill Cornell Medical College; institutional review board review was not required as data was deidentified.

Results

The probability of a PLCO trial participant undergoing a biopsy as a function of the maximum PSA value from all tests increased at the 4.0 ng/mL PSA cutoff by 27.3% (95% CI, 23.3%-31.3%; P < 1 × 10−10) (Figure). This translates into a relative 445% increase in the biopsy rate for those with a PSA just above 4.0 ng/mL compared with those just below that cutoff.

At a PSA of 4.0 ng/mL, biopsy based on screening increased the absolute detection rate of low-risk (Gleason score ≤6 at clinical stage T1-2a) prostate cancer by 7.2% (95% CI, 3.6%-10.8%; P = 8.5 × 10−5) (Figure and Table). There was no effect on the detection of intermediate-risk (Gleason score = 7 or clinical stage T2b) (P = .94) (Table) or high-risk (Gleason score ≥8 or clinical stage T2c-3a) (P = .98) (Figure and Table) prostate cancer.

Examining the pathology from those who underwent prostatectomy yields similar results. There was a discontinuity in the detection of cancers with a Gleason score of 6 or lower (5.6% [95% CI, 2.6%-8.7%]; P = <.001) and no discontinuity in the detection of scores of 7 (P = .52) or 8 to 10 (P = .56) (Table). We found no discontinuity in prostate cancer–specific mortality (P = .27) or overall mortality (P = .62) (Figure and Table).

Discussion

Using RD in the screening arm of the PLCO trial, we were able to effectively instrument for biopsy based on PSA screening. Despite excluding the control arm of the study, we confirm the study results of the PLCO trial that biopsy at a PSA threshold of 4.0 ng/mL did not decrease prostate cancer–specific or overall mortality. For prostate cancer–specific mortality, the trend was toward increased mortality at the cutoff, which is opposite of what would be expected if there were a benefit to screening.

Interpreting clinical data using RD has many potential applications in medicine since treatment decisions are often based on discrete cutoffs in continuous data.3

Back to top
Article Information

Corresponding Author: Jonathan Shoag, MD, Department of Urology, New York Presbyterian Hospital Weill Cornell Medical College, 525 E 68th St, Starr 900, New York, NY 10021 (Jes9171@nyp.org).

Published Online: August 20, 2015. doi:10.1001/jamaoncol.2015.2993.

Author Contributions: J. Shoag and D. Shoag had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: J. Shoag, Eisner, Lee, Mittal, D. Shoag.

Acquisition, analysis, or interpretation of data: J. Shoag, Halpern, Lee, Barbieri, D. Shoag.

Drafting of the manuscript: J. Shoag, Halpern, Barbieri.

Critical revision of the manuscript for important intellectual content: J. Shoag, Halpern, Eisner, Lee, Mittal, Barbieri, D. Shoag.

Statistical analysis: J. Shoag, Halpern, Lee, D. Shoag.

Administrative, technical, or material support: Lee.

Study supervision: J. Shoag, Eisner, Lee, Barbieri.

Conflict of Interest Disclosures: Dr Eisner works as a consultant for Percys, Boston Scientific, Olympus, Cook, Radius Pharmaceuticals, Bard. Dr Eisner is also an owner of Ravine Group. No other conflicts are reported.

Additional Information: We thank the National Cancer Institute for access to data collected by the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial. The statements contained herein are solely those of the authors and do not represent or imply concurrence or endorsement by the National Cancer Institute.

Correction: This article was corrected on August 27, 2015, to fix a figure error.

References
1.
Andriole  GL, Crawford  ED, Grubb  RL  III,  et al; PLCO Project Team.  Mortality results from a randomized prostate-cancer screening trial.  N Engl J Med. 2009;360(13):1310-1319.PubMedGoogle ScholarCrossref
2.
Andriole  GL, Crawford  ED, Grubb  RL  III,  et al; PLCO Project Team.  Prostate cancer screening in the randomized Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial: mortality results after 13 years of follow-up.  J Natl Cancer Inst. 2012;104(2):125-132.PubMedGoogle ScholarCrossref
3.
Moscoe  E, Bor  J, Bärnighausen  T.  Regression discontinuity designs are underutilized in medicine, epidemiology, and public health: a review of current and best practice.  J Clin Epidemiol. 2015;68(2):122-133.PubMedGoogle ScholarCrossref
4.
Imbens  GW, Lemieux  T.  Regression discontinuity designs: a guide to practice.  J Econom. 2008;142(2):615-635.Google ScholarCrossref
5.
Imbens  K, Kalyanaraman  K.  Optimal Bandwidth Choice for the Regression Discontinuity Estimator.http://www.nber.org/papers/w14726. 2009. Accessed July 14, 2015.
6.
Nichols  A.  RD: Stata module for regression discontinuity estimation. http://ideas.repec.org/c/boc/bocode/s456888.html. 2011. Accessed July 7, 2015.
×