Population-Based Breast Cancer Screening in a Primary Care Network
Published Online: December 18, 2012
Steven J. Atlas, MD, MPH; Jeffrey M. Ashburner, MPH; Yuchiao Chang, PhD; William T. Lester, MD, MS; Michael J. Barry, MD; and Richard W. Grant, MD, MPH
Patient characteristics, mammography reports, and dates of completion were obtained from an electronic clinical and billing data repository.15 Physician characteristics were obtained from the hospital registrar.
The primary outcome was mammography completion rates among patients overdue for screening at the start of the study (prevalent cohort) and among women newly overdue for mammography (incident cohort) during the first study year, comparing intervention and control practices.12 The maximum length of follow-up was 3 years for those in the prevadlent cohort and at least 2 years among those in the incident cohort.
A mammogram was considered to have been completed if there was an electronic report for an imaging test at a network-affiliated institution or if a mammogram was listed in billing data for the patient. Secondary outcomes included time to mammography completion among all overdue patients (prevalent and incident cohorts), censored by cancer diagnosis, death, or end of follow-up. New cancer diagnoses using Partners Healthcare. Cancer Registry data were compared among intervention and control groups.
Baseline patient and physician/practice characteristics were compared between intervention and control groups and between prevalent and incident cohorts using 2-sample t tests or x2 tests, as appropriate. For the primary outcome, adjusted mammography completion rates and 95% confidence intervals were calculated for both the prevalent and incident cohorts at 1, 2, and 3 years of follow-up using Cox proportional hazard models with the robust sandwich covariance matrix estimate to account for clustering while adjusting for potential confounders (PROC PHREG in SAS version 9.2; SAS Institute Inc, Cary, North Carolina). In these models, physicians were considered as the unit of cluster for physician-connected patients and the population manager was considered the unit of cluster for practice-connected patients. To control for differences in patient and practice characteristics among intervention and control practices, patient age, race/ethnicity, insurance status, English language proficiency, practice type (health center vs non–health center), and number of months since last practice visit were included in the models as covariates. All adjusted rates were calculated by holding these covariates at the population mean levels. Unadjusted time to screening completion survival distributions were depicted with Kaplan-Meier curves and compared using a log-rank test. Adjusted hazard ratios comparing intervention with control practices for the entire follow-up period were also reported from the Cox proportional hazards models. The percentages of patients with new cancer diagnoses were compared between intervention and control groups using x2 tests.
There were 64 eligible physicians and 6 practice population managers in the 6 intervention practices and 74 eligible physicians in the 6 control practices. Among intervention providers, 65 of 70 (92.9%) used the system. There were no significant differences between intervention and control practice physicians with regard to age (47.4 vs 46.9 years; P = .78), years since medical school graduation (19.9 vs 19.2; P = .67), or sex (48.4% vs 51.4% were women; P = .86). Two practices in each arm were community health centers. Screening rates at baseline were similar in intervention and control practice groups (79.5% vs 79.3%; P = .73). Figure 1 depicts practice randomization and follow-up.
Among 32,688 eligible women, 9795 (30%) were overdue for screening during the 1-year study period, including 4487 patients in intervention practices and 5308 patients in control practices (Table 1). Intervention and control patients were equally likely to be connected to a specific physician (58.9% vs 58.8%) and overdue for screening at the start of the study (prevalent cohort, 67.9% vs 68.8%). Intervention patients were slightly younger, and more likely to speak English, to be non-Hispanic white, to have health insurance, and to have their last clinic visit further in the past than control patients. Compared with patients who were overdue at the start of the study (prevalent cohort), patients who became overdue during 1-year follow-up (incident cohort) were more likely to be connected to a physician (67.6% vs 54.8%; P <.001), to have commercial health insurance (70.9% vs 61.9%; P <.001), and to have been seen more recently for a practice visit (mean 8.6 [8.6 standard deviation (SD)] months vs 13.8 [8.1 SD] months; P <.001).
Providers took action on 3415 of 4487 (76.1%) intervention patients; 2865 (63.9%) were contacted by letter and 550 (12.3%) were deferred. The most common reasons for deferral included test completion at an outside facility (312 [56.7%]) and patient refusal (89 [16.2%]). Among 3045 intervention patients overdue at baseline (prevalent cohort), action was taken on 2629 (86.3%) patients; 2212 (72.6%) were contacted by letter and 417 (13.7%) were deferred. Among 1442 intervention patients who became overdue during the 1-year follow-up period (incident cohort), action was taken on 786 (54.5%) patients; 653 (45.3%) were contacted by letter and 133 (9.2%) were deferred.
Mammography Screening Rates Over Time Percentage of Overdue Population Screened Over 3-Year Follow-up. The percentage of patients in the intervention and control groups who completed screening over the 3-year follow-up period was analyzed separately for prevalent and incident cohorts. Among patients overdue at baseline (prevalent cohort), adjusted completion rates were significantly higher among patients in the intervention group compared with the control group at 1 year (30.1% vs 26.0%; P = .004), 2 years (41.5% vs 36.2%; P = .002), and 3 years (51.7% vs 45.8%; P = .002) of follow-up (Figure 2A). Among patients becoming overdue during the first year (incident cohort), adjusted completion rates were higher but of borderline statistical significance in the intervention group compared with the control group at 1 year (39.8% vs 35.5%; P = .07) and 2 years (53.8% vs 48.7%; P = .052) of follow-up (Figure 2B).
PDF is available on the last page.