Effectiveness and Cost of an Enhanced Mailed Fecal Test Outreach Colorectal Cancer Screening Program: Findings from the PROMPT Stepped-Wedge Trial

Screening can substantially reduce colorectal cancer incidence and mortality: (1, 2) Of the 50,000 annual U.S. deaths from colorectal cancer, up to 30,000 (60%) could be averted by routine screening (3). Screening participation is suboptimal in the United States, especially among adults who receive care at community health centers. In 2021, only 42% of adults ages 50 to 75 served by community health centers were up to date with colorectal cancer screening recommendations (4). Moreover, 2021 data from the National Health Interview Survey show that a smaller share of Hispanic adults ages 45 to 75 (52%) were up to date with colorectal cancer screening recommendations than non-Hispanic White adults (61%; ref. 5). Mailed fecal immunochemical test (FIT) outreach programs can improve colorectal cancer screening participation, with a meta-analysis reporting an improvement of 28 percentage points (6–12). Typically, these programs send patients a mailed FIT kit with information about screening, along with advance notifications (before the kit is received) and/or reminders (after the kit is received) in the form of letters, telephone calls, or text messages (6, 7, 10–12).

Despite widespread use of advance notifications and reminders in mailed FIT outreach, data are sparse on how these communications affect screening rates. One recent meta-analysis of four studies reported colorectal cancer screening improvements with use of advance notification letters compared with no advance notification, with a pooled relative risk of 1.09, 95% confidence interval (CI; 1.07–1.11; refs. 13–17). A second meta-analysis (four studies; two overlapped with the first) reported a median screening improvement of 4.1 percentage points [interquartile range (IQR; 3.6%–6.7%)] for advance notification by letter or automated phone call compared with no advance notification (14, 17–20). While these analyses show promise for advance notifications, five of the six included studies evaluated an advance notification letter; only one evaluated automated phone calls (reporting a 4.5 percentage point increase in screening), and none evaluated text messages (19).

Similar gaps exist in literature on benefits of mailed FIT reminders. Two studies found that live phone call reminders following a mailed FIT boosted FIT completion by 2.7 to 3.5 percentage points (18, 21, 22). In a pilot study, our team found a 3-month FIT completion of 23.7% with a mailed FIT and reminder letter. FIT completion dropped by 0.4 and 6.8 percentage points, respectively, when the letter was replaced with an automated phone call or text message. However, FIT completion rose when a live reminder call was delivered, either alone (+8.1 percentage points) or in combination with a letter (+3.7 percentage points), automated call (+5.2 percentage points), or text message (+3.4 percentage points; ref. 23). However, the sample size for this individual randomized study was relatively small; as a result, there is limited robust evidence of the effectiveness of low-cost advance notifications or reminders.

We compared the effectiveness of a standard mailed FIT outreach program that included live reminder calls with an enhanced program that also included advance notifications and automated phone call reminders. The study used a stepped-wedge, cluster-randomized design to facilitate the pragmatic roll out of the program by clinic staff. We included an embedded patient-level randomization that compared advance notification live phone calls with text messages in patients with no electronic health record (EHR) evidence of prior colorectal cancer screening. The program was administered in fifteen clinics within a large, urban federally qualified health center network. We assessed effects on FIT completion and overall colorectal cancer screening rates at 6 months. We report implementation and cost of each program component.

The study setting, rationale, and design have been reported previously (24–26). Briefly, the study used an open cohort, cluster-randomized, stepped-wedge design and was conducted at a subset of 15 clinics (clusters) within a large federally qualified health center in Southern California. The health center operates 25 medical clinics and served over 270,000 patients in 2018, including 41,050 patients ages 50–75. More than 97% of patients ages 50 to 75 listed English (39%) or Spanish (59%) as their preferred language. The health center uses the InSure ONE FIT (Clinical Genomics), which requires collection of two specimens from a single stool sample. FITs are processed at a reference laboratory (Quest Diagnostics). The health center reported that 63% of eligible patients were up to date with colorectal cancer screening in 2018.

All procedures and intervention materials were reviewed and approved by the Kaiser Permanente Northwest Institutional Review Board (IRB no. 1394084-36:32), with ceding agreements from the health center and Oregon Health & Science University (Portland, OR). The study obtained a waiver of informed consent and authorization for use of protected health information, given the minimal risk posed to patients.

We used boot camp translation (BCT; ref. 27), a validated community engagement process, to develop the content and select the format of advance notifications and reminders for the enhanced program (25). The in-person BCT sessions (one in English and one in Spanish) informed content for advance notification text messages, advance notification live phone call scripts, and automated phone call reminders. During sessions, we also solicited input on patient preferences regarding mode (text messages, automated or live phone calls) and timing of advance notifications and reminders. All materials were developed in English and Spanish and reviewed and approved by the health center's marketing department (Supplementary Table S1).

We pilot-tested three versions of the program in two health center clinics, using an individual randomized design. In addition to a FIT, participants were randomized to receive: (i) a text message advance notification and two automated phone reminders (automated condition); (ii) up to three live phone reminders (live condition); or (iii) a text message advance notification, two automated phone reminders, and up to three live phone reminders (combined automated plus live condition). Findings have been reported previously (28). Briefly, 26% of adults in the automated condition completed a FIT within 3 months, significantly fewer than in the live [32.3%; +6.3%; 95% CI (1.1–11.4)], or the combined automated plus live [35.7%; +9.7%; 95% CI (4.4–14.9)] conditions. Subgroup analysis showed that screening-naïve patients (no prior colorectal cancer screening completed) had lower FIT completion rates than patients who had completed prior colorectal cancer screening [17.2% vs. 45.1%; OR, 3.97; 95% CI (3.19–4.95); ref. 24].

We selected automated plus live outreach (enhanced mailed FIT outreach) for the main trial because it achieved the highest FIT completion in the pilot study, and compared it with live outreach alone, as this was the usual care practice at the health center (standard mailed FIT outreach). Given the low FIT completion among screening-naïve patients, we also included an embedded patient-level randomization to live phone call advance notifications or text message advance notifications in patients with no EHR evidence of prior colorectal cancer screening. We hypothesized that screening-naïve patients may be unaware of the need for screening and thus more likely to respond to an advance notification delivered by live phone call than by text message.

The main trial involved sequential crossover of clinics (i.e., clusters) from usual care to intervention. We included all primary care clinics operated by the health center that had more than 200 age-eligible patients (n = 19), except for clinics with planned closures (n = 2), or that participated in the pilot (n = 2). On the basis of prior research (29), we stratified clinics based on colorectal cancer screening rate (>40%; ≤40%) and the proportion of patients who listed Spanish as their preferred language (>70; ≤70%), then the project statistician randomized them into two wedges. Wedge 1 clinics crossed over from the standard mailed FIT program to the enhanced mailed FIT program in June of 2018; wedge 2 clinics crossed over in June of 2019. All clinics were randomized on May 31, 2018.

For the main trial, centralized outreach staff used a list of EHR codes developed in previous studies (30) to identify adults who had attended at least two clinic visits within the past 24 months and were overdue for colorectal cancer screening (31). Across the baseline, year 1, and year 2 time periods, we identified 27,585 instances where these criteria were met at the 15 participating clinics (7,101 at baseline, 8, 31 in year 1, and 11,553 in year 2). Some adults met criteria in more than one time period. Clinic staff in a centralized outreach department mailed FIT kits to all qualifying adults in each time period.

Within each wedge, and immediately prior to intervention roll-out, patients who had no EHR evidence of having completed prior colorectal cancer screening (1,749 in year 1 among wedge 1 clinics and 3,047 in year 2 among wedge 1 and wedge 2 clinics) were randomized to receive either advance notification by text message or by a live phone call delivered by a member of the centralized outreach team. Patient-level randomization occurred at two timepoints in year 1 (June 22, 2018, n = 1,358; September 24, 2018, n = 391 non-managed care Medicaid patients) as non-managed care Medicaid patients were inadvertently omitted from the first year 1 mailing, and at one timepoint in year 2 (June 11, 2019, n = 3,047).

For practical reasons, neither researchers nor centralized clinic outreach staff were blinded to clinic-level or patient-level randomization assignment. The intervention was delivered by clinic staff; eligible adults were unaware of their study participation consistent with the approved IRB protocol.

Patients in the standard mailed FIT condition were mailed a FIT kit with an introductory letter and a one-page educational sheet developed through the BCT process (26). One month following the mailing, patient lists were “refreshed” to remove patients who had completed their FIT and needed no further reminders. One to 2 months following the mailing, live reminder calls were made during clinic hours (Monday through Friday) by care team members (usually medical assistants). Care team members were bilingual in English and Spanish, and interpreter services were available for patients who spoke other languages. Care team members made up to three reminder attempts, leaving voicemail messages when patients did not answer; patients who were reached or whose contact information was incorrect received no further reminder attempts. To encourage patient response, the outgoing phone number for live calls was a local clinic number. FIT kit packets were mailed on July 11, 2017 in the baseline year, July 9, 2018 and October 15, 2018 in year 1, and July 8, 2019 in year 2 (Supplementary Table S2).

In year 1, eligible adults in clinics assigned to the enhanced mailed FIT outreach condition were delivered the same components as the standard mailed FIT outreach on the same schedule, as well as advance notifications (either by text message or live phone call) and additional automated call reminders. Centralized outreach staff were trained to deliver advance notification calls by project staff during a 4-hour in-person session. All patients who had records of completing a prior FIT (n = 12,135) were sent advance notification text messages; individuals who had never completed a prior FIT (n = 4,794) were randomized to receive either a text message or live phone call (Supplementary Table S2) in advance of receiving the FIT in the mail. Text messages were sent 2 to 3 days after FIT kits were mailed; 1–2 days before the patient was expected to receive the kit. Advance notification live phone calls were delivered starting 3 weeks before and up to the anticipated receipt of the FIT by centralized outreach staff. Adults whose FITs were not returned within 1 week were sent two automated phone call reminders, delivered 11 and 22 days after the FIT mailing date. Automated calls reminded patients to complete the FIT and invited patients to press “0” to be transferred immediately to a staff member at the health center's Patient Contact Center if they had questions. Texts and automated calls were sent by a contracted vendor (Stericycle Communication Solutions); the outgoing phone number was a local number. Following the automated calls, care team medical assistants made up to three attempts to reach patients for a live phone call reminder, delivered on the same schedule as the standard mailed FIT outreach condition. In year 2, all participating clinics received the enhanced mailed FIT outreach.

We describe the demographic characteristics of all study participants across each intervention condition, as well as the percentage of FITs completed within 6 months of patient identification. We report unstandardized effect sizes and associated 95% CIs for all analyses.

Our primary intention-to-treat analysis (ITT; all clinics and participants evaluated in their assigned wedge) compared FIT completion rates across treatment conditions using the generalized linear mixed model described by Hussey and colleagues (32). The model included treatment status (control vs. intervention) and time (categorical: baseline, 1 year, 2 years) as fixed independent variables and clinic as a random effect. We repeated this analysis for our secondary endpoint of any colorectal cancer screening within 6 months of patient identification.

We performed per-protocol (PP) analyses for both outcomes by limiting the sample to patients known to have received the program in each condition; the first PP analysis (PP1) included those who were successfully delivered an advance notification live call (patient answered or voicemail message left) or text message (message sent) as well as received the automated call (patient answered, voicemail message left, or unknown whether person or voicemail received call); the second PP analysis (PP2) was further limited to individuals who were known to have answered the advance notification live call and the automated call.

Finally, we examined whether baseline age (50–64 vs. 65–74), sex (male vs. female), ethnicity (Hispanic vs. non-Hispanic), preferred language (English/Spanish or other), attendance at a clinic visit in the past year (no visits vs. 1 or more visits), and history of a FIT (no prior FIT vs. at least 1 prior FIT) modified the treatment effect. We performed each subgroup analysis using the same stepped-wedge generalized linear mixed model framework for the primary (FIT completion within 6 months) and secondary (any colorectal cancer screening within 6 months) outcomes by adding the moderator and product of moderator and arm. We tested for moderation in the ITT and PP samples.

To test whether live phone call advance notifications led to higher rates of FIT completion than text message advance notifications in the subgroup of patients with no EHR evidence of prior colorectal cancer screening, we used a generalized linear mixed model where notification type (0 = text message, 1 = live call) was the independent variable and clinic was modeled as a random effect. To investigate whether time moderated this effect, we expanded the model by including time (year 1, year 2) and the product of time and notification type (representing the interaction) as independent variables. These analyses were repeated for the secondary outcome of any colorectal cancer screening within 6 months.

Given the small number of time periods, we were unable to test the assumption that the effect of the enhanced outreach is constant for all clinics across time (33). However, we examined the trajectory for each wedge using the same model as the primary and secondary outcomes, but including wedge assignment (wedge 1 vs. wedge 2) instead of intervention status, time, and the product of time and wedge as independent variables.

We report the number of patients who were sent each component of the intervention (text message or live call advance notification, automated phone call reminder) for year 1 clinics (eight clinics) and year 2 clinics (seven additional clinics).

We estimated costs for each component of the enhanced mailed FIT program. We asked project staff to record the time they spent on each component; these estimates were used to derive labor costs, with inflations of 20% to cover overhead costs and 30% to cover fringe benefits. We used actual wage rates when possible and national wage rates (https://www.bls.gov/oes/current/oes211094.htm) when needed. Staff time estimates were requested after the program was operational to most closely represent steady-state costs. Because the intervention was delivered as part of a research study which may have included additional, non–intervention-related costs, we adjusted the estimates to subtract protocol-driven costs. This allowed us to estimate replication costs (i.e., costs that might be incurred after adopting the intervention without major modification; ref. 34).

The data generated in this study are not publicly available due to data use agreements for the study; but are available upon reasonable request from the corresponding author.

A total of 15 clinics participated: eight in wedge 1 and seven additional in wedge 2 (Fig. 1). There were 7,101 patients at baseline (all received standard care), 8,931 at year 1 (5,380 in clinics allocated to enhanced intervention, 3,550 in clinics allocated to standard care), and 11,553 at year 2 (all allocated to enhanced intervention). Patients had a mean age of 59 years (SD = 6.0), 57% were female and 82% were Hispanic (Table 1). Eighty-four percent had visited the clinic at least once in the previous year (SD = 3.5) and 68% listed Spanish as their preferred language. Two-thirds of the patients had past FIT completion documented in the EHR, and nearly 71% had evidence of at least one prior-year no-show for a scheduled visit. The patients at baseline, year 1, and year 2 had similar demographics and health care use characteristics (Supplementary Table S3).

Overall, 37% of patients who were mailed a FIT completed one within 6 months. We observed post-intervention increases in both 6-month FIT completion (+2.8%; 95% CI, 0.4–5.2; Table 2) and in any colorectal cancer screening completion (+2.5%; 95% CI, 0.1–5.0) in ITT analysis.

We observed larger effect sizes in our PP analysis for FIT completion (PP1: +4.0%; 95% CI, 1.0–6.9; PP2: +16.9%; 95% CI, 12.9–21.0) and any colorectal cancer screening (PP1: +3.8%; 95% CI, 0.9–6.7; PP2: +16.3%; 95% CI, 12.3–20.3).

We observed evidence of heterogeneity in treatment effects for having a clinic visit in the past year [interaction OR = 0.64; 95% CI (0.51–0.81); P < 0.001]. Those without a prior year clinic visit had a higher boost in 6-month FIT completion (+24.9%) compared with those with a prior year visit [+15.0%; difference-in-change = 9.9%; 95% CI (4.5%–15.3%)]. All other tests of moderation were nonsignificant.

Among those with no EHR evidence of prior colorectal cancer screening, the advance notification live call was marginally superior to text message for FIT testing (+2.2%; 95% CI, 0.0–4.5) but did not reach significance for any colorectal cancer screening completion (+2.0; 95% CI, −0.4 to 4.3; Table 3). However, time had a significant moderating effect on FIT completion: live call was superior to text message for FIT completion at year 1 (+ 6.4%; 95% CI, 2.2–10.6) but not at year 2 (+0.5%; 95% CI, −2.2 to 3.3). We found no significant interaction of time and wedge (Supplementary Fig. S1).

A total of 9,468 patients completed a FIT: 4,208 in the preintervention phase and 6,260 in the postintervention phase. Of these, 670 (6.4%) had an abnormal test result; 205 of those (30.6%) had a colonoscopy within 12 months of their result (chart abstraction was performed on a subset of 231 in year 1; ref. 35).

A total of 27,585 patients were mailed a FIT kit: 7,101 at baseline, 8,931 in year 1, and 11,553 in year 2 (Table 4). All patients were delivered a text or phone call advance notification. All patients allocated to the text message condition were sent a text; implementation of the automated phone call reminders ranged from 91% to 96% of eligible patients. Across all years, usual care delivery of live reminder calls by members of the care team was relatively low, ranging from 14% to 27% of intended recipients. These calls were made at the discretion of medical assistants who were also performing clinical care.

Costs were similar for the standard mailed FIT group ($10.48 per patient, inclusive of FIT and mailing costs) and those in the enhanced mailed FIT group who were sent text message notifications ($11.14 per patient; Table 5). Costs for those in the enhanced mailed FIT group who were sent live call notifications were nearly double that of the other groups ($21.30 per patient; $10.82 per patient additional cost) due primarily to the costs of delivering the live call ($8.07 per patient), and training outreach staff ($2.22 per patient).

Adding advance notifications and automated reminders to a mailed FIT outreach program with live phone call reminders led to an improvement in FIT completion rates of about 3 percentage points [+2.8%; 95% CI (0.4–5.2)] in an urban, primarily Hispanic/Latino-serving community health center. The effect rose to 17 percentage points [+16.9%; 95% CI (12.3–20.3)] in an analysis limited to patients who were successfully reached by advance notifications and reminders. Beyond the initial FIT mailing costs ($10.84 per patient), automated intervention components required minimal additional costs (+$.66 per patient). In contrast, the addition of automated plus live outreach (live phone calls plus automated call reminder) doubled the cost (+$10.82 per patient). In a subanalysis of patients with no prior FIT testing history, live advance notification phone calls improved FIT completion by 6 percentage points over text messaging in year 1, but no improvement was observed in year 2.

A recent meta-analysis of FIT outreach programs found that advance notifications and reminders led to modest incremental improvements in screening, with median absolute increases of 4.1% (IQR, 3.6%–6.7%) and 3.1% (IQR, 2.9%–3.3%), respectively (18). This is consistent with our observed 3 percentage point increase in screening completion, an effect that rose to 17 percentage points in analysis limited to patients known to have received the study components (i.e., received text message and answered phone call outreach).

Our findings have implications for Hispanic patients, who traditionally have lower colorectal cancer screening rates than non-Hispanic White patients. Consistent with prior research by our team and others, we observed overall higher FIT completion among Hispanic than non-Hispanic adults (39.5% vs. 31.8%; ref. 23). This finding corresponds with a preference for FIT over colonoscopy among Hispanic adults who are given the choice between two tests, as reported by Inadomi and colleagues (36). Further research is needed to assess the reach and effectiveness of outreach modalities among Hispanic adults, given their relatively low participation in colorectal cancer screening. Ongoing efforts also are needed to assure that those with abnormal FIT results obtain a follow-up colonoscopy, as these rates are low in the general population (56% completion within 1 year of abnormal test result; based on national health insurance and health record data; ref. 37), and particularly low in health center settings (range = 18%–57% completion within 6 months of abnormal test result, based on chart abstracted health record data; ref. 38).

Our current data show that this intervention effect was attenuated in year 2 data. The attenuated effectiveness corresponded with an overall drop in colorectal cancer screening rates at the health center from 2018 to 2020. Uniform Data System data show that the overall colorectal cancer screening rate of eligible adults seen at the health center was 63% in 2018, and fell to 56% in 2019 and 43% in 2020. This might be explained, in part, by a change in EHR provider (from NextGen to Epic) that occurred in 2019. During this time, fewer patients were offered FIT during in-clinic visits, due to clinic staff training and delays in reinstating EHR functionality to identify patients at point-of-care who were due for screening. The drop in FIT offers during in-clinic visits led to more patients being eligible for the mailed outreach program; we observed expected increases in the number of eligible patients from 7,101 in 2017 to 2018 (baseline year) to 8,931 in 2018 to 2019 (year 1) and 11,553 in 2019 to 2020 (year 2). In-clinic FIT offers generally result in a higher FIT completion than mailed FIT outreach; thus, a drop in in-clinic offers would have led to lower overall screening rates, as we observed.

Notably, our year 2 evaluation interval ended in December 2019, prior to the onset of COVID-19. Thus, the drop in year 2 effectiveness was not impacted by COVID-19–related care suspension. In contrast to the large drops in health center screening rates observed from 2018 to 2020, we found only moderate variation across years in FIT completion rates among patients mailed a FIT. These rates ranged by 6.3%, from 35.3% to 41.6% (baseline: 41.6% in wedge 1 vs. 40.4% in wedge 2; year 1: 39.9% in wedge 1 vs. 36.3% in wedge 2; year 2: 35.8% in wedge 1 vs. 35.3% in wedge 2) and diminished incrementally across program years.

Our findings underscore the important role of centralized outreach teams in delivering advance notifications and reminders; we observed higher implementation of advance notification live phone calls by a centralized team (100% of calls delivered, 70% of patients reached), than live phone call reminders delivered by decentralized care teams (14%–27% of calls delivered). Finally, we report a low cost associated with automated outreach, suggesting that these modest improvements could be achieved with minimal additional costs. Moreover, a prior evaluation reported costs of mailed FIT and live phone call outreach of $22 to $24 per patient, consistent with our estimated per-patient cost of $21.30 (39). Finally, our findings from the subanalysis of patients with no prior FIT testing history, may have direct implications for the 22 million 45 to 49 year olds who became screening eligible with the 2021 updated U.S. Preventive Services Task Force guidelines; for this group, live advance notification phone calls might improve FIT completion (over text messaging); but more research is needed on how to maintain that improvement over time.

Our study had several strengths, including its large and diverse sample, randomized design, and the near-complete capture of FIT events and demographic characteristics in the EHR. Moreover, we applied a validated engagement process and used pilot testing to design and refine intervention materials and components. Our pragmatic study design and evaluation mean that our findings should generalize to real-world practice settings.

The study also had some limitations. It was difficult to correctly classify receipt of intervention components: automated phone messages can be delivered to unintended recipients, and we could not be certain that delivered text messages were read. Also, though a stepped-wedge design lends itself well to pragmatic trials, the design makes assumptions that the effects of time are constant across clinics; though we performed sensitivity analyses to examine this assumption, interpretation was limited by the number of wedges and evaluation intervals (32). In addition, because fecal testing is recommended annually, some individuals participated at multiple timepoints. Finally, our analysis was limited to a 2-year timespan.

Adding live and automated advance notifications and automated phone call reminders to a standard mailed FIT outreach program with live reminder phone calls moderately boosted FIT completion and any colorectal cancer screening rates. These enhancements could be added for low to moderate additional per-patient costs.

G.D. Coronado reports grants from NIH during the conduct of the study; grants from Guardant Health and personal fees from Exact Sciences outside the submitted work. D.B. Nyongesa reports grants from NIH during the conduct of the study. J.H. Thompson reports grants from National Institute on Minority Health and Health Disparities during the conduct of the study. D. Smith reports grants from NIH during the conduct of the study. M.M. Davis reports grants from NCI during the conduct of the study. M.C. Leo reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

G.D. Coronado: Conceptualization, investigation, writing–original draft, writing–review and editing. D.B. Nyongesa: Formal analysis, writing–review and editing. A.L. Escaron: Writing–review and editing. A.F. Petrik: Project administration, writing–review and editing. J.H. Thompson: Project administration, writing–review and editing. D. Smith: Data curation, investigation, writing–review and editing. M.M. Davis: Writing–review and editing. J.L. Schneider: Data curation, writing–review and editing. J.S. Rivelli: Data curation, writing–review and editing. T. Laguna: Supervision, writing–review and editing. M.C. Leo: Writing–review and editing.

This research was funded by a grant from the National Institutes of Minority Health and Health Disparities (U01 MD010665). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The NIH had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the article; and decision to submit the article for publication.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.