The STOP CRC study Auto and Auto Plus interventions were successfully implemented in two safety-net clinics. Both interventions led to higher colorectal-cancer testing rates than rates in the usual care clinic, demonstrating the effectiveness of an EHR-embedded intervention addressing colorectal-cancer screening. Our pilot findings showed high reach for the mailed component (based on the low number of mailed items that was returned from the Post Office), and moderate reach for the phone-call component (based on 2 call attempts). Further research is needed to assess effectiveness of the program as an on-going part of standard clinical care (not as a one-time mailing), and to assess the adoption, implementation, and maintenance of the program. If successful, the program may represent an effective method of raising levels of participation in colorectal-cancer screening and improving earlier-stage detection of colorectal-cancer among patients least likely to be screened.
Our findings showed substantially higher colorectal-cancer testing rates in our two interventions clinics, compared to similar patients in a third VGMHC clinic that did not receive the intervention. The differences in rate of fecal testing in our two intervention sites versus the usual care site (difference in Auto Clinic vs. Usual care: 38% and difference in Auto Plus Clinic vs. Usual care: 35%) were higher than effect sizes observed in previous clinical studies on the same topic [4–7].
Our point estimate for differences in fecal testing rates between our Auto and Auto Plus clinics was marginal (Difference in differences: 38% - 35% = 3%). This may be due, in part, to the lower response in the Auto Plus clinic to the mailing of the introductory letter, and reminder postcard (FIT return rate: 32%), compared to the Auto Clinic (39%). Of the 66 Auto Plus patients identified for theory-based phone counseling, 8% of those identified, and 17% of those successfully reached, returned their FIT kits. Pooling our FIT completion rates for the 2 clinics, our best estimate of effectiveness of the Auto intervention alone is 36%, plus another 2% from phone-based follow-up. This is consistent with findings from 3 studies that used telephone reminders or theory-based phone counselling [4, 15, 16], but differed from a study conducted by Green et al. at Group Health Cooperative, which showed an added bump of 7 percentage points associated with brief phone assistance, and a further bump of 7 percentage points with more intensive ongoing phone-based navigation . It is important to note that Green et al. used medical assistants and/or nurses who were hired by the study to deliver the interventions, whereas STOP CRC integrated intervention delivery into routine care. We cannot rule out the possibility that the apparent lack of effect of the phone counseling in our pilot was due to small sample sizes or differences in baseline characteristics of clinics or selected patients.
Our observation that only 16/213 (7.5%) participants were found to have an invalid address (as determined by their introductory letter or kit being returned by the Post Office) was contrary to expectation. This may be due, in part, to a system-wide mailing to update patient address information that took place 3 months before our introductory letter was sent. Notably, while we observed high reach for our mailed components, it is plausible that some mailings were not received by their intended participants. Also, we anticipate that clinics with less up-to-date patient address information will achieve lower reach.
While our sample size is too small to permit statistical comparisons across subgroups, our pilot data are suggestive of high levels of effectiveness among Hispanics and other individuals who speak Spanish. Notably, among Auto Clinic patients, the highest rate of fecal testing was found among those who had 6 or more clinic visits; this suggests that personal interactions with a provider in addition to the mailed program may serve to reinforce the importance of screening. This finding is consistent with data from Liles et al. in a study that enrolled patients at Kaiser Permanente Northwest .
Our pilot program has some limitations that we plan to address in the larger multi-site study. Our inclusion and exclusion criteria rely on EHR data, and we could not verify the accuracy of colonoscopy receipt, raising the possibility that our intervention was delivered to patients who were ineligible due to recent colorectal-cancer screening. Nevertheless, a minority of patients opted out (n = 8), and only 3 opted out because of prior testing. We plan to address this by conducting a robust validation of EHR codes used for our inclusion and exclusion of participants for the larger study. We also plan to enhance the capture of colorectal-cancer screening in EHR-based tools for tracking outside screening events (called Health Maintenance in Epic). Our feasibility assessment relied on quantitative data only; we plan to report separately on feasibility considerations based on qualitative interviews with providers and patients. Moreover, we report no data on the cost of providing affordable testing and follow-up care for patients in this setting, which may drive feasibility and sustainability over time.
The small size and non-random nature of our sample limit the interpretation of our findings. Intervention effects are inextricably confounded with clinic effects, and the interventions were delivered only to patients in the practices of a single team (2–3 providers and their support staff of a registered nurse, patient care coordinator, and team assistant processing referrals) in each clinic. The patient panels appeared to differ with regard to the proportions that were excluded because of prior colorectal-cancer screening and other factors. These providers volunteered for the intervention and may have been more willing to involve their staff in conducting follow-up calls than providers in the clinic as a whole. Nevertheless, because the 3-sample gFOBT cards, and not the FIT, were offered during clinic encounters as part of usual care, we could easily discern that our findings were not impacted by more frequent recommendations for screening during clinic encounters. Nevertheless, the differences in screening probabilities between intervention and usual-care clinics were striking and we will use them to help inform power calculations for the larger study.
Our pilot provided some important information that will inform the design of a large-scale pragmatic study to test the effectiveness of the program in multiple safety-net clinics. We report successful implementation, high reach for mailed components, moderate reach for telephone components, and high effectiveness for both interventions. We were also able to successfully embed our registry tools into the EHR, and use real-time data to identify patients eligible for each intervention step.
These findings, as well as findings from on-going analysis of qualitative interviews with patients and providers, will inform several aspects of a planned multi-clinic study that will enroll a broad range of patients. Specifically, our preliminary estimates of effectiveness suggest that additional telephone-based outreach may not be needed. Further exploration of how a variety of factors may influence preventive services use may be needed to inform further refinements to the program.