A Population-based Randomized Trial to Promote Cancer Screening Among Unscreened Medicaid Recipients in Minnesota

This trial was conducted between April 2014 and July 2015 and implemented through Sage, the National Breast and Cervical Cancer Early Detection Program (NBCCEDP) in Minnesota and housed within the Minnesota Department of Health (MDH). The target population was all Minnesota Medicaid recipients ages 50-74 overdue for breast cancer and colorectal cancer screening. Claims data, obtained from MDHS, were used to determine patient characteristics and outcomes. Individuals who were not enrolled in Medicaid, who were not overdue for BC or CRC screening, or who were not in the 50-74 age range were excluded prior to analysis.

To determine the efficacy of the interventions we used a two group posttest-only randomized design with all eligible MA beneficiaries randomly assigned to one of two groups: Direct Mail plus Incentive (with patient navigation) versus usual care.

The primary outcomes were completion of mammography or colonoscopy within 12 weeks after implementation of the intervention. Current Procedural Technology (CPT) codes from Medicaid claims data were used to determine if an individual had received either mammography or colonoscopy. Specific CPT codes used for mammography were conventional mammography (77055-77057), digital mammography (G0202, G0204, G0206), and computer-aided detection mammography (77051, 77052). CPT codes used for colonoscopy were G0105, G0121, 45378, and 45380-45385. Mammography and colonoscopy outcomes were measured dichotomously. Outcomes were based on the presence of screening mammography and colonoscopy claims occurring for 10 weeks after the first mailings.

The mammography and colonoscopy interventions were examined separately. Across treatment and control, study sample characteristics were compared and absolute differences were assessed using t-test and χ2 statistics. Main outcome analyses consisted of logistic regression to compute odds ratios for receiving mammography or colonoscopy and for covariate adjustment. Two separate logistic regression models were examined for both mammography and colonoscopy interventions: (1) a bivariate model that examined treatment versus control, and 2) a multivariate model that adjusted for covariates. All analyses were conducted using Stata, version 13.

Some individuals lost Medicaid coverage on a monthly basis during the study period, and others had inaccurate mailing addresses (< 2% of each study sample). Therefore not everyone received the treatment as intended, and individuals that received the intervention may not have had Medicaid claims available post-intervention. Our outcome analyses were intent-to-treat analyses that included all individuals randomized to treatment or control groups at initiation of intervention regardless of whether they lost coverage post-randomization or if they had an inaccurate mailing address.