Our analysis is based on data from the 2000 and 2003 National Health Interview Surveys (NHIS) and the National Cancer Institute's (NCI) 1999–2000 Survey of Colorectal Cancer Screening Practices (SCCSP), Primary Care Physician Questionnaire (12). The NHIS, conducted by the National Center for Health Statistics of the Centers for Disease Control and Prevention (CDC), is nationally representative and collects information about demographic characteristics, chronic health conditions, health insurance, and health behaviors of the civilian, noninstitutionalized U.S. population (www.cdc.gov/nchs/nhis.htm). The NHIS Cancer Control Supplement, administered in 2000 and 2003, includes a series of questions about the use of CRC screening (16).

The SCCSP, conducted by NCI, CDC, and the Centers for Medicare and Medicaid Services, surveyed a nationally representative sample of 1235 practicing primary care physicians for 1999–2000, including family and general practitioners, general internists, and obstetricians and gynecologists (8,12). The survey was designed to estimate CRC screening capacity and the knowledge and beliefs of primary care physicians about CRC screening. Details of the sampling scheme and a description of the characteristics of the respondents have been published (12).

We merged data from the NHIS with data from the SCCSP at the county level. These data were then aggregated to hospital-referral regions, which represent regional health care markets for medical care and have been used extensively to examine regional variation in health care use (17-24). The United States has 306 hospital-referral regions (17). We constructed independent variables to reflect the recommendations of primary care physicians in the region where each respondent lived. Because of the confidential nature of these data, analyses were conducted at the Research Data Center of the National Center for Health Statistics.

We included data on individuals from the NHIS who were aged at least 50 years, had not previously received a diagnosis of CRC, and responded to questions regarding the use of CRC screening. Because the sampling frames of the NHIS and the SCCSP were not identical, we limited our sample to individuals who lived in a hospital-referral region where four or more primary care physicians were surveyed in the SCCSP (N = 12,727 individuals in 122 hospital-referral regions).

NHIS respondents were asked several questions about their use of CRC screening: if they had ever had an FOBT using a home test kit and, if so, the timing of their most recent home FOBT; if they had ever had a CRC screening test by sigmoidoscopy or colonoscopy and, if so, the type of test and the timing of their most recent test. Individuals were classified as "ever screened" for colorectal cancer if they reported ever taking a home FOBT, having had a sigmoidoscopy, or having had a colonoscopy. We also examined whether subjects who had reported CRC screening were current on screening, (i.e., home FOBT during the past year, sigmoidoscopy within the past 5 years, or colonoscopy within the past 10 years [2]). NHIS did not ask about barium enema. Although the American Cancer Society includes double contrast barium enema every 5 years as an acceptable screening option (3), the U.S. Preventive Services Task Force did not find direct evidence that this method is effective in reducing CRC mortality (2).

We used data from the NHIS to define individual characteristics and included age, sex, race and ethnicity, education, health insurance, health-care-seeking behavior, prior history of cancer other than CRC, number of chronic medical conditions, and number of behavioral risk factors for CRC. We categorized ethnicity as non-Hispanic white, non-Hispanic black, Hispanic, or other race and ethnicity. Educational attainment was defined as less than high school graduation, high school graduate, some college, and college graduate. Health insurance categories were uninsured; Medicare with private supplemental insurance, or private insurance; Medicare without supplemental coverage; and Medicaid or dual eligibility for Medicare and Medicaid. We categorized health-care–seeking behavior according to whether an individual had a usual source of health care, evidenced by a visit to any health care professional, including a dentist, in the past year. Chronic medical conditions included arthritis, peptic ulcer disease, chronic lung disease, cardiovascular disease, hypertension, and diabetes. Behavioral risk factors for CRC included current cigarette use, heavy drinking (consuming 60 or more alcoholic drinks per month for men and 30 or more for women), and lack of regular exercise (25-27).

Primary care physicians who participated in the SCCSP were asked at what age and how frequently they recommended each CRC screening method for a patient at average risk (12). Physicians reporting the use of FOBT were asked whether they provided office-based or home tests. Because sensitivity is lower for a single office-based FOBT than for the home test, in which samples are collected over 3 days (2,4), only a home test was considered adequate screening. We coded recommendations for initiation of each type of CRC screening as being in accordance with screening guidelines if the physician reported recommending that patients begin having the test at age 50 (2). We coded each physician's belief about frequency of screening as being in accordance with the guidelines if the response was at the recommended interval for at least one type of CRC screening test. We aggregated all responses according to hospital-referral region to create two region-level measures of CRC screening practices: 1) the proportion of primary care physicians in a hospital-referral region who recommended initiating CRC screening at age 50 years, and 2) the proportion of these physicians who advised at least one screening test at the recommended interval. These two variables captured distinct information supported by a correlation coefficient of only −0.04.

Using the data on the individual as the unit of analysis, we constructed multilevel logistic regression models to examine the odds of undergoing CRC screening. We based the models on the average proportion of primary care physicians in an individual's hospital-referral region who recommended CRC screening, after controlling for individual factors associated with CRC screening. To reflect the greater precision of estimates from hospital-referral regions with a large number of primary care physicians responding, we adjusted NHIS survey sample weights for the number of primary care physicians per region. The odds ratios (OR) for region-level measures of physician recommendations were expressed for a 30-percentage-point increase in the proportion of primary care physicians in the region recommending CRC according to the guidelines. Models accounted for the clustering of individuals in regions and for the survey sample weights and were estimated with SAS 9.0 (SAS Institute, Cary, North Carolina). We based independent variables on prior work and on their statistical relationships with the dependent variable. We hypothesized that regional physician belief about the age at initiation of CRC screening would be associated with the likelihood that an individual living in a region would ever be screened. We also hypothesized that regional physician recommendations about screening intervals would be associated with the likelihood that a patient would be current on screening and included this variable in this model. The final models included age, sex, race and ethnicity, education, insurance, usual source of care, prior diagnosis of cancer other than CRC, dental visit within the prior year, number of chronic health conditions, number of behavioral risk factors for CRC, year of NHIS survey, and relevant hospital-referral region measure.

Only 50.2% of adults aged 50 years or older had ever been screened for CRC (Table 1). Hispanics were significantly less likely than non-Hispanic whites to have been screened. Respondents with less than a college degree were less likely than college graduates to have been screened. Uninsured respondents, those who had Medicare without supplemental coverage, and those with Medicaid or who were dually eligible for Medicare and Medicaid were less likely than those with private insurance or Medicare plus a supplemental policy to have ever been screened. Respondents without a usual source of care were less likely than those with one to be screened. Respondents who had previously received a diagnosis of cancer were more likely than those with no diagnosis to have been screened. CRC screening increased with the number of chronic conditions, but decreased as the number of behavioral risk factors for CRC increased. CRC screening increased between 2000 and 2003.

On average, 53.3% of primary care physicians in a hospital-referral region recommended initiating CRC screening at age 50 (range 0%–100%). In regions where less than 30% of physicians recommended initiating screening at age 50, 47.3% of respondents had been screened, in contrast to 54.8% of respondents in regions where 70% or more of physicians made this recommendation. After adjustment for individual characteristics, an absolute increase of 30 percentage points (e.g., from 50% to 80% or 20% to 50%) in the proportion of primary care physicians in a hospital-referral region who recommended initiating CRC screening at age 50 was associated with a higher prevalence of screening in that region (OR, 1.09; 95% CI, 1.01-1.18).

Among respondents who had ever received CRC screening, 77.9% were current on screening (Table 2). Women were less likely than men to be current on screening. Respondents who had some college education were less likely than those who had graduated from college to be current on screening. We found no association between current CRC screening and race and ethnicity, insurance, prior diagnosis of cancer other than CRC, the number of chronic health conditions, or the number of behavioral risk factors for CRC. Respondents without a usual source of care were less likely than those with one to be current on screening. The proportion of respondents who had been screened and were current on screening increased between 2000 and 2003.

On average, 64.8% of primary care physicians in hospital-referral regions recommended at least one CRC screening test at the recommended interval (range 0%–100%). In regions where <30% of physicians advised screening at the recommended frequency, 70.7% of respondents were current on screening, in contrast to 79.9% of respondents living in areas where ≥70% of physicians made this recommendation. After adjustment, an increase of 30 percentage points in the proportion of primary care physicians in a hospital-referral region who recommended at least one CRC screening test at the correct interval was associated with a higher prevalence of current screening in that region (OR, 1.19; 95% CI, 1.05-1.37).

Seventy-one percent of physicians who correctly indicated that screening should begin at age 50 reported recommending at least one test at the correct interval. Of these physicians, 54.4% recommended initiating screening at age 50. Overall, 37.4% of physicians correctly recommended both initiation and frequency.