Study population
The American National Cancer Institute instituted the PCOS in 1994 to measure practice patterns and health-related quality of life among men diagnosed with prostate cancer in the United States. Methods for this multi-site, longitudinal project are described elsewhere [11]. Briefly, PCOS subjects were men histologically diagnosed with prostate cancer between October 1, 1994 and October 31, 1995. Subjects were identified using a rapid case ascertainment system by the six participating National Cancer Institute Surveillance, Epidemiology and End Results (SEER) cancer registries (Atlanta, Georgia metropolitan area; Los Angeles County California; King County, Washington; Connecticut; Utah; and New Mexico). Eligible subjects were residents of the areas covered by these registries at the time of diagnosis and were between the ages of 39 and 89 years, except in King County, where only men over 60 years were eligible. The institutional review board of each participating institution approved the study.
Eligible patients were sampled within strata of age, race/ethnicity, and tumor registry to approximate a sample representative of the United States population of prostate cancer patients. The PCOS oversampled younger men and minorities and excluded patients with race/ethnicity other than non-Hispanic white, African American, or Hispanic, because their sample sizes were small.
A total of 11,137 men with prostate cancer comprised the eligible patient population for the study and the PCOS randomly selected 5,672 of these men. Among these selected patients, 3173 (55.9%) completed a health-related quality-of-life survey questionnaire 6 months after initial diagnosis. We used survey and medical record data collected from these subjects to evaluate differences in patient characteristics and treatments between men with screening-detected cancers and those who were diagnosed clinically. Responders to the PCOS survey were younger than non-responders and more likely to be non-Hispanic white and have a higher socioeconomic status. A substantial proportion of the responders had regional stage and moderately differentiated cancers, while non-responders had a greater proportion of distant stage and poorly differentiated cancers. Responders also were more likely to receive radical prostatectomy [11].
Data collection
Investigators contacted eligible subjects by mail and/or telephone requesting them to sign a release form allowing review of all medical records from any physicians and facilities diagnosing and/or providing care for prostate cancer. Records were obtained from private and public hospitals, freestanding radiological or surgical centers, Veterans Administration hospitals, Health Maintenance Organizations, and private physician offices. Certified Tumor Registry abstractors collected baseline information on demographics, clinical symptoms before diagnosis (systemic and urinary), comorbidity, diagnostic procedures and results (including PSA levels and digital rectal examination findings), clinical staging, tumor characteristics, and treatment details. The PCOS re-abstracted a random sample of 5% of records to assess and correct any systematic coding errors.
The PCOS also collected data on general and disease-specific measures of health-related quality of life, symptoms, comorbidity, and specific treatments received for prostate cancer using a mailed self-administered questionnaire. Most respondents completed the self-administered questionnaire (91%); those who did not return the questionnaire were contacted by telephone and asked to complete the survey by telephone or in person. Subjects were asked to recall their health-related quality of life and symptoms, including the domains of urinary, bowel, and sexual function, just before their prostate cancer was diagnosed. Demographic and socioeconomic questions from this survey were used to determine race/ethnicity, employment status, educational level, household income, insurance status, and marital status. A question assessing comorbidity asked about 12 medical conditions that were likely to affect prostate cancer treatment decisions and long-term quality of life. The conditions were derived from the Charlson index as well as the expert opinion of the PCOS investigators [12]. If the patient reported being told by a doctor that he had cerebrovascular disease, inflammatory bowel disease, liver disease, or ulcers, he received one point on his comorbidity score for each condition. If the patient reported that any of eight conditions – arthritis, diabetes, depression, hypertension, chest pain, heart attack, heart failure, or chronic lung disease – limited his activity or required prescription medications, he received 1 additional point for each of these conditions. In the analyses, comorbidity scores were divided into the categories of 0, 1, 2, and greater than or equal to 3 points.
We assigned screening status using information from the medical record abstract and the patient questionnaire. We considered men presenting with symptoms consistent with advanced prostate cancer, including bone pain, weight loss or hematuria, to be "clinically diagnosed." We initially created separate categories for men with only irritative or obstructive symptoms consistent with benign prostatic hyperplasia and an asymptomatic group who had neither prostate cancer nor lower urinary tract symptoms.
Clinical cancer stage was based on an algorithm using information abstracted from medical records. The algorithm was necessary because the community-based medical records were not detailed enough to classify cases by TNM (tumor, node, metastases) staging [13]. The algorithm defined T1 tumors as confined to the prostate with a normal digital rectal examination and no positive scans (magnetic resonance imaging, computed tomography, bone scan) or evidence of metastases. T2 tumors were defined as confined to the prostate, with abnormal or suspicious digital rectal examinations, but no positive scans or evidence of metastases. We defined clinically localized cancers as either T1 or T2 tumors. Initial treatment, based on medical record abstractions, was defined as treatment received within the first six months after diagnosis. We defined aggressive treatment as either radical prostatectomy or radiation therapy. We defined conservative management as androgen deprivation, either surgical or chemical, or watchful waiting.
Statistical analyses
Descriptive statistics were calculated for ethnicity/race, age, stage at diagnosis, education, marital status, employment, income, digital rectal exam and PSA results, Gleason score from biopsy or transurethral resection of the prostate, comorbid conditions and self-reported general health status. We used contingency tables to compare men presenting without any symptoms, those with lower urinary tract symptoms alone, and those with prostate cancer symptoms. Although screening is defined as applying a diagnostic test to asymptomatic people [14], the prevalence of benign prostatic hyperplasia is very high among men at risk for prostate cancer [15]. We found that the men with only lower urinary tract symptoms were much more similar to asymptomatic men than to men we classified as having clinically diagnosed cancers. Therefore, we also considered cancers diagnosed in men who reported only lower urinary tract symptoms at the time of PSA testing to be "screening-detected." We used this combined screening-detected group to compare baseline characteristics against clinically diagnosed cases and in modeling treatment selection for clinically localized cancers. Logistic regression analyses were used to determine whether screening history was independently associated with selecting aggressive treatment versus conservative management among men with clinically localized prostate cancer. Covariates for this multivariate model, based on previous literature, included age, race/ethnicity, marital status, study site, education, insurance status, annual income, comorbidity, health status, and tumor characteristics [16, 17]. We also examined interactions between screening status with age, comorbidity, PSA level, and Gleason score.
The results of the logistic regression models are shown as percentages receiving the treatment of interest, adjusting for the independent variables included in the model. These percentages were directly adjusted to the distribution of the variables among the weighted sample used in each model [18]. The probability of receiving the treatment of interest can then be directly compared across levels of the variables included in the model.
All analyses were performed with the Survey Data Analysis statistical package (Research Trial Institute, Research Triangle Park, North Carolina, 1997) to account for the complex survey design. We obtained unbiased estimates of parameters for all eligible prostate cancer patients in the PCOS areas by using the Horvitz-Thompson weight, which is the inverse of the sampling proportion for each sampling stratum (defined by age, race/ethnicity, and study area). A two-tailed P-value of < .05 was considered statistically significant.