This study was conducted as part of the NCI-funded Population-based Research Optimizing Screening through Personalized Regimens (PROSPR) consortium.18 The overall aim of PROSPR is to conduct multi-site, coordinated, trans-disciplinary research to evaluate and improve cancer screening processes. In 2014, we conducted a survey of PCPs affiliated with the three primary care networks of the PROSPR breast cancer research centers (response rate 57.6%) to ascertain provider beliefs about the effectiveness of mammography and their recommendations for screening.11 Resident physicians were excluded. These data were linked to the screening utilization data for the patients of participating PCPs for the period from 2011 to 2014, obtained from PROSPR’s central data repository. All sites received institutional review board approval for active or passive consent processes or a waiver of consent to enroll participants, link data, and perform analyses.
Settings and Study Population
Women were eligible for inclusion in the study if they were 40–89 years of age, without a known diagnosis of breast cancer or ductal carcinoma in situ, and had a primary care visit between January 1, 2011, and September 30, 2012, with a PCP affiliated with the clinical network of Brigham and Women’s Faulkner Hospital (BWFH; Boston, MA), Dartmouth-Hitchcock Health System (DH; Lebanon, NH), or the University of Pennsylvania (PENN; Philadelphia, PA). This time frame was chosen to ensure that most women had 2 years of follow-up in the data repository. One of the provider networks (PENN) was limited to only 18–24 months of follow-up to observe screening mammograms after the PCP visit.
Data and Covariates
Provider-level data were obtained from the survey, which ascertained information about provider demographic and practice characteristics, including age, gender, specialty (family or general practice, general internal medicine, gynecology, physician assistant, nurse practitioner, certified nurse midwives, other), medical school affiliation, number of outpatient office visits in a typical week, number of physicians at their practice, whether during the prior 12 months they had their clinical income adjusted based on their breast cancer screening performance, whether they had ever been sued for failure to diagnose cancer, and whether their primary practice had received NCQA recognition as a medical home.
Also derived from the survey were the principal independent variables: provider beliefs concerning screening mammography effectiveness, and their recommendations for screening frequency. Survey questions about screening beliefs and recommendations were assessed by specific age groups (i.e., 40–49 years, 50–74 years, and ≥75 years), because guideline recommendations for screening vary by patient age. For each age group, PCPs were asked the following: (1) How effective do you believe screening mammography is for reducing cancer mortality for average-risk women? (response options: very effective, somewhat effective, not effective, effectiveness not known, I am not sure); and (2) How often do you recommend screening mammography for most asymptomatic, average-risk women (in good health for their age)? (response options: every year (12 months), every other year (24 months), I discuss interval options with patients, I do not routinely recommend).
Patient-level data were obtained from PROSPR’s central data repository including age (categorized as 40–48 years, 49–74 years, and ≥75 years to reflect guideline differences by age; 49-year-olds were included with the older group because they turned 50 before the end of the 2-year window), race/ethnicity (white, black, Hispanic, Asian/Pacific Islander, American Indian, mixed race/other), type of health insurance (commercial, Medicare, Medicaid, uninsured/medical assistance, other, or missing), and Charlson comorbidity score (0, 1, 2+).19
Survey data were linked to data from PROSPR’s central repository using a unique provider identification code. Of the 73,266 women in the PROSPR data repository meeting the eligibility criteria for this study, 30,233 (41.2%) linked to a PCP who had completed a survey.
Outcome variables were obtained from PROSPR’s central data repository and measured at the patient level, as follows: (1) whether a woman received a screening mammogram during the 2 years following a PCP visit; and (2) for women who received a screening mammogram, whether they received >1 mammogram during that period, reflecting the use of annual screening. A screening mammogram was defined as a mammogram coded by the radiology facility as being for screening; in addition, the woman did not have any prior imaging in the prior 3 months, to avoid inclusion of diagnostic mammograms.
We performed logistic regression analyses of the two binary outcomes. Potential predictors were assessed using univariate and then multivariate analysis, adjusting for other significant factors. For the assessment of patient characteristics, we used a population-averaged generalized estimating equations (GEE) model to account for correlation among outcomes from the same PCP. To evaluate provider characteristics and beliefs we used GEE models adjusted for patient characteristics found to be significant in the patient-level GEE analyses. All models of provider characteristics and beliefs were adjusted for patient age, race/ethnicity, and insurance status as well as calendar year and primary care network (BWFH, DH, PENN). Additionally, for evaluating the first outcome of “any screening use” we adjusted for comorbidity. Significant provider variables were tested in a larger joint multivariate model. The absolute measurement of screening may have been reduced at PENN due to shorter follow-up than the other two sites. However, we performed a sensitivity analysis excluding PENN (14% of the population), and there were no substantial differences. Thus, we report results for all three sites.