We used 2007–2009 data from the linked Surveillance, Epidemiology, and End Results (SEER)–Medicare database,3
cancer incidence, and survival from patients in geographic areas representing 28% of the US population, cross-matched with the Medicare enrollment master file, along with a 5% sample of noncancer Medicare beneficiaries residing in SEER program areas.
Using a quasiexperimental design, we compared longitudinal changes in PSA-based prostate cancer screening among men 75 years and older with concurrent screening trends among men aged 66 to 74 years as a control group, a difference-in-differences approach. By using a multiple time series with a comparison group, the approach reduces bias from unmeasured variables and from secular trends. Our prerecommendation and postrecommendation periods were 15 months from April 2007 through June 2008 and October 2008 through December 2009, respectively, which allowed a brief “washout” period for dissemination of the August 2008 USPSTF recommendation. Consistent with prior research,4
PSA-based prostate cancer screening was determined using Healthcare Common Procedure Coding System codes. Men screened multiple times during a period were only counted once.
Because wide regional variation in prostate cancer screening and treatment has been demonstrated,5
we subsequently examined whether there was a differential impact of the 2008 USPSTF recommendation among hospital referral regions (HRRs) that varied in prerecommendation PSA-based screening rates among men 75 years or older and urologist density.6
For analytical purposes, HRRs were categorized as having low (first quartile), medium (second and third quartiles combined), and high (fourth quartile) prerecommendation screening rates and urologist density.
We used a generalized linear model that included observation period (prerecommendation vs postrecommendation), age (66–74 years vs ≥75 years), and an interaction between these 2 variables, along with race and Elixhauser comorbidity score, to estimate the differential impact of the 2008 USPSTF recommendation. These analyses were then repeated to examine whether the differential impact of the recommendation varied across HRRs stratified by both prerecommendation PSA-based prostate cancer screening rates and urologist density. All analyses were conducted using SAS version 9.2 (SAS Institute Inc).