We found that the majority of cancers (89.3%) in a population-based PCOS cohort were detected by screening. Compared to men with clinically diagnosed prostate cancer, men with screening-detected cancers were younger, more likely to be married, less likely to be a member of a minority group, and in better health. The cancers detected by screening were more likely to be clinically localized and less likely to be poorly differentiated. Among men with clinically localized prostate cancers, those with screening-detected cancers were significantly more likely to undergo aggressive treatment, even after adjusting for demographics, comorbidity, and tumor characteristics
Our finding that a high proportion of prostate cancers diagnosed in 1994 and 1995 were detected by screening is consistent with the temporal correlation between the increased use of PSA testing and the increased incidence of prostate cancer in the USA beginning during the early 1990s [4
]. Although prostate cancer incidence rates decreased for several years in the mid 1990s, more recent data show that incidence rates are again increasing [5
] and survey results from the Centers for Disease Control's Behavioural Risk Factor Surveillance System (BRFSS) show that a high proportion of American men continue to undergo PSA testing [20
]. These data suggest that our findings are still relevant for prostate cancers being diagnosed in the USA. We also found that men with screening detected cancers were more likely to have early stage cancers, again mirroring the epidemiologic data showing an increased incidence of early stage cancers and a decreased incidence of advanced stage cancers [4
]. The majority of screening-detected tumors were moderately to poorly differentiated; however, a significantly higher proportion of clinically diagnosed cancers were poorly differentiated.
Previous data, including an analysis of the PCOS cohort, have shown African Americans to be twice as likely as non-Hispanic whites to present with advanced stage cancers [4
]. In the current analysis, we found a greater prevalence of ethnic/racial minorities in the clinically diagnosed versus screening-detected cancers. This disparity may reflect ethnic/racial differences in accessing preventive health care services, particularly arising from socioeconomic barriers. This in turn could contribute to disparities in cancer stage at diagnosis [22
]. However, African Americans also have been reported to demonstrate more skeptical attitudes towards screening [25
] and the stage disparity could be due to racial differences in tumor aggressiveness [26
Men with screening-detected clinically localized cancers were more likely to undergo aggressive treatment with radical prostatectomy or radiation therapy than men with clinically diagnosed cancers. The odds ratio for receiving aggressive treatment was statistically significant at 1.5, but the adjusted absolute difference between screening-detected and clinically diagnosed cases was only 6 percentage points. This modest association between screening status and treatment selection suggests that clinical practice may be only partly consistent with the American College of Physicians' view that "aggressive treatment is necessary to realize any benefit from the discovery of a tumor [27
]." Our findings may reflect the scientific uncertainty about whether and how to treat screening-detected prostate cancers [28
Our study has some potential limitations. We classified men presenting with symptoms of advanced cancer as being clinically diagnosed. We do not know that these symptoms actually prompted diagnostic PSA testing. However, the tumor registry medical record abstractors are trained to identify the events leading to a cancer diagnosis; they would attempt to record only symptoms consistent with cancer. Classifying PSA as a screening test is also difficult given the high prevalence of lower urinary tract symptoms in older men [15
]. Few members of our study cohort were truly asymptomatic because nearly two-thirds reported lower urinary tract symptoms. However, our classifications for clinical diagnosis and screening detection were internally valid because men diagnosed with symptoms of advanced cancer were significantly more likely to present with advanced stage and more aggressive cancers than the combined group of men who were either asymptomatic or had only lower urinary tract symptoms. Additionally, when we compared demographic and socioeconomic characteristics across groups, we generally found that the men with lower urinary tract symptoms alone most closely resembled the asymptomatic men.
Selection bias may have occurred because 44% of the sampled patients did not complete the 6-month survey. Responders were younger than non-responders, more likely to be non-Hispanic white, had higher socioeconomic status, had earlier stage disease, and were more likely to receive radical prostatectomy. Results may be less generalizable to older men, those with lower socioeconomic status, or members of racial/ethnic groups other than non-Hispanic white. However, these were also the groups who were less likely to have screening-detected cancers. We do not believe that including these non-responders would have altered our findings on the differences between screening-detected and clinically diagnosed cancers. However, based on their demographics, socioeconomic status, and advanced disease stage, the non-responders were not likely to have a high proportion of screening-detected cancers and thus we may have overestimated the proportion of screening-detected cancers. Another potential limitation arose from asking subjects to recall their baseline symptoms 6 months after diagnosis. Recall errors could lead us to misclassify screening status. However, Legler and colleagues prospectively studied a subset of PCOS subjects and found high concordance for symptom recall at 6-months after diagnosis compared with reports at the time of diagnosis [29
]. Finally, we may have had incomplete symptom data, particularly for questions appearing only in the medical record abstract. The abstracts would report a symptom if it appeared in the medical records; the absence of a symptom could be due to either the patient being asymptomatic or the physician's failure to ask about or record the symptom. We performed a sensitivity analysis by using only subject reported symptoms from the survey, then only symptoms reported on the medical record abstract, and then ultimately using a combination of both sources. The results for all analyses were essentially the same.