The health insurance plan is a non-profit organization that provides insurance in Oregon for Medicaid, Medicare, and dental coverage for about 220,000 enrollees at the time of the study. The mailed FIT program was implemented in collaboration with a research team in 2016 (May to November) and 2017 (May to November). A total of 17 health systems took part in the BeneFIT program offered by the Oregon health insurance plan over the first two years that the program was implemented. We report here on 2-year CRC screening rates and implementation variations (i.e., adaptations) for these health systems. Six of these health systems had the capacity to provide FIT test results to the research team and implement the program quickly enough to be included in a prior analysis of the one-year mailed FIT outcomes.27
Mailed FIT Intervention
The BeneFIT program is described in detail elsewhere.20 Briefly, health plan staff generated lists of enrollees due for CRC screening for each health system that took part in the program. To be eligible for the mailed FIT program, a member must have been between the ages of 51 and 75 and not have had a health plan claim indicating CRC screening or a screening exclusion (i.e., colon cancer). Health plan staff provided the member lists and FIT kits to a mail vendor that prepared and mailed introductory letters. Enrollees whose introductory letter was returned as undeliverable were removed from the list. The mail vendor mailed remaining enrollees a FIT kit about 4 weeks later, followed by a postcard reminder two weeks later.
Within this framework, each health system was allowed by the health plan to customize how they implemented the program. For all health systems, the FIT results came back directly to clinics and were followed up directly by the patient care teams using the clinics’ usual care procedures. The basic program (specifically the mailing elements coordinated by the health plan) was presented to clinic managers, who then determined if they would be able to add clinic-supported adaptations, such as phone call reminders. The adaptations (i.e., differences in implementation) fell into five types:
- Lists of eligible enrollees scrubbed before mailing the introduction letters: Health systems could review the list of eligible members that the health plan generated and remove patients based on their own patient data. Health system staff either looked for patients who were current for screening according to clinic-based medical records or simply validated that the patients correctly belonged to the clinic’s population [e.g., were regularly seeing one of their providers or had an electronic health record (EHR)]. The health system then returned a “scrubbed” list back to the health plan.
- 12-month visit exclusion: Some clinics chose to have the health plan automatically exclude patients who had not had a clinic visit in the last year. In this case, the health plan staff removed patients without a visit in the last 12 months using the claims database. (Often this adaptation was chosen simply because clinics could not staff the effort of scrubbing the mailing lists.)
- Phone call reminders: Some health systems had staff call patients who were mailed an introduction letter and FIT kit to remind them to return the test. The health plan provided the clinic staff with a list of plan members who were mailed an introduction letter and FIT kit.
- Financial incentives (gift card) offered for completing CRC screening: In some regions or health systems, the health plan offered incentives for completion of CRC screening (either by FIT or by colonoscopy). The incentives ($25 gift cards) were mentioned in the letters that accompanied the FIT kits.
- Allowing FIT kits to be mailed back (vs. requiring in-person drop off): Some health systems required members to return the completed FIT kits in person to a clinic. Other health systems allowed members to mail back the completed kits in pre-stamped return mailers that were provided when the kits were sent (referred to as a mailed return).
In addition to these five major implementation variations, other health care system characteristics were available for the analysis. Some of the health systems had participated in prior research efforts involving mailed FIT outreach and therefore had some existing FIT mailing workflows and staff experience. The health systems varied in size, both in number of clinics and number of patients they served. Finally, the health plan allowed the program to mail whichever type of FIT was already in use by the health system. All health systems used one of the following three types of FIT: the two-sample Insure® by Clinical Genomics, one-sample Hemosure® by Hemosure, Inc., or one-sample OC-Light® or OC-Auto® by Polymedco.
The main outcome for these analyses was completed CRC screening rates. A screening was considered complete if a claim was submitted that indicated a patient received any type of CRC screening procedure within six months of the date that the introductory letter was mailed. A CRC screening procedure was defined as any of the following:
- FIT test or fecal occult blood test (FOBT)
- FIT-DNA test
- Flexible sigmoidoscopy
- Computed tomography (CT) colonography (virtual colonoscopy)
The number of FIT kits mailed indicates the number of eligible health plan members who were mailed a FIT kit through the BeneFIT program. All implementation outcomes were tracked internally by health plan staff as they generated lists of eligible patients and worked with health systems and the mailing vendor to conduct the mailing itself.27 For FIT kits mailed in late 2017, there was a minimum three-month period for claims to be received by the health plan following the six-month screening period.
Each variable was a potential explanatory factor that could have a plausible connection to the outcome. Health plan characteristic variables included health system name, health system size (number of clinics per system), participation in the prior CRC screening study, and FIT test type used by the health system. Intervention variables included the length of participation in BeneFIT, number of adaptations, number of kits mailed, list scrubbing, 12-month visit exclusion, reminder calls, patient incentive, and a mailed return option.
This study incorporated a multi-method approach. A descriptive analysis comparing CRC screening completion rates by health system characteristics and interventions was completed using Minitab and Tableau Software. Configurational Comparative Methods (CCMs) analyses were then performed using the R package “cna” to analyze the dataset using Coincidence Analysis (CNA).29-31 RStudio, R, and Microsoft Excel were also used to support the configurational analysis with CCMs. The configurational analysis examined the combinations of adaptations and health system characteristics that together distinguished the health systems with higher screening rates from those with lower screening rates.
The configurational analyses used a dichotomous outcome for each of three analyses: Percent Completed Year 1, Percent Completed Year 2, and Change from Year 1 to Year 2 (positive or not positive). We set the threshold for our main outcome, CRC screening completion rate, at 19%. This cutoff was determined by tertiles, where we compared cases in the upper two tertiles of the screening rate percent versus cases in the lowest tertile. In the Year 1 analysis, there were 17 cases in the overall sample, with 11 cases in the upper two tertiles and 6 cases in the lowest tertile. For Year 2, there were 10 cases in the overall sample, with 7 cases in the upper two tertiles and 3 cases in the lowest tertile. In both Year 1 and Year 2, the 19% cut point separated the upper two tertiles from the lowest tertile, and in both years there was a sizable performance gap in the outcome across this threshold, a difference of more than 3.5 points in absolute terms. Only health systems that participated in both years were included in the Year 2 analysis and the change from Year 1 to Year 2 analysis.
The configurational analyses produced an overall model with high consistency and coverage that identified combinations of conditions that explained the presence of the outcome. Consistency refers to how often health systems identified by the model had the outcome present (i.e., higher screening rates); coverage accounts for the percent of health systems with higher screening rates explained by the model.
To achieve data reduction, we used a configurational method to identify candidate factors, described in detail in prior studies.32-34 To summarize, we used the “minimally sufficient conditions” function within the R package “cna” to look across all 17 cases and all 8 factors at once. The consistency threshold was initially set to 100% and the coverage threshold to 15%. We considered all 1-, 2-, 3-, 4- and 5-condition configurations in our dataset that met this dual threshold. If no configurations met these criteria during the data reduction phase, we iteratively dropped the consistency threshold by increments of 5 percentage points (i.e., from 100% to 95%) and repeated the process of creating a new condition table until configurations emerged that satisfied all criteria.
Next, we sorted the condition table by complexity and coverage, and identified the configurations with the highest coverage scores. We began with 1-condition configurations to see if they met the consistency and coverage thresholds and were uniquely distinguished from all other 1-condition configurations. We then proceeded to examine 2-, 3-, 4- and 5-condition configurations, working upwards to minimize possible redundancy. Using this approach, we reduced the dataset to a smaller subset of candidate factors. We selected final solutions based on high overall model consistency (i.e. as close to 100% as possible, and at least 80%) and coverage (i.e., as close to 100% as possible, and at least 70%).