All study activities were approved by Mass General Brigham Institutional Review Board. The study was registered on Clinical Trials.gov (NCT03959696). Reporting followed the CONSORT guidelines.11 This work was funded by the Patient-Centered Outcomes Research Institute (PCORI) grant CDR-2017C3-9720.
The PRIMED study is a multi-site cluster randomized trial that enrolled primary care physicians and assigned them to Intervention or Comparator arms. Patients were surveyed after a visit with a participating physician to assess outcomes.
Primary care physicians from Internal Medicine and Family Medicine practices affiliated with five hospital networks—three academic medical centers and two community hospitals in the Northeast—were eligible if they had at least 20 patients aged 75–85 due for a CRC screening or surveillance in their panel. Residents and advance practice providers were excluded because, in most of these hospital networks, they do not carry their own patient panels.
The study staff extracted a list of potentially eligible patients from each participating physician’s panel. Physicians reviewed their lists and excluded any patient based on the criteria in Table 1. Research staff reviewed physicians’ schedules to identify potentially eligible patients with an upcoming preventive care or scheduled follow-up appointment and confirmed eligibility before inviting them to participate.
SDM Skills Training
A 2-h online SDM communication skills training course included case studies and interactive exercises to simulate conversations with older patients about CRC testing (see Table 2 for details). The course was adapted from previous training sessions12,13 and incorporated features that have been shown to improve the effectiveness of continuing medical education (CME).14 Physician participants received 2 h of CME. They also received resources to use throughout the study, including a patient-facing education worksheet, the ability to submit cases and get feedback from study investigators, and an opportunity to complete an additional telephone-based simulated patient interaction to practice skills. Five email newsletters were sent to intervention arm physicians (about one per quarter during patient enrollment) summarizing key points from the challenging cases that were submitted by physician participants.
Research staff sent an email or electronic health record (EHR) message (per physician preference) to the participating physician 2–3 days prior to a visit with an eligible patient. The message encouraged the physician to have a conversation with the patient about whether to continue CRC testing and included the patient’s last CRC test and date, if applicable.
Randomization and Blinding
Eligible physician participants were grouped into strata based on self-reported gender, years in practice, prior exposure to SDM training (self-reported by physicians at enrollment), and site. Within each stratum, we assigned physicians to one of two arms (Intervention or Comparator) using a random number generator. It was not possible to blind physician participants to study interventions; however, no details were given regarding the content of the arms (only an estimate of the time required for each arm). Physicians were aware that their patients would be surveyed but were not given details on the content of the survey. Patients were blinded to the study arms. The PIs and the project manager were not blinded to physicians’ assignment as they needed to arrange access to training courses. Research staff who entered the patient survey data into REDCap and the biostatistician conducting the analyses were blinded.
We enrolled physicians from May 2019 through August 2019. Physicians were notified of the study at practice meetings, presentations, newsletters, and individual invitations. Interested physicians were screened to confirm eligibility and indicated consent by sending an email confirming their intention to join.
From October 28, 2019, through March 13, 2020, eligible patients were mailed a study invitation and an information sheet describing the study 2–3 weeks before a visit. The cover letter included information for patient participants to opt out by calling or sending a postcard to the study team. Shortly after the scheduled visit, staff confirmed that the visit occurred and sent a packet including a $5 incentive to all patients who had not opted out. The staff made up to three reminder calls and sent a reminder packet to patient non-responders about 4 weeks after the initial packet. Due to the low-risk nature of the study, patient consent was implied by returning the survey.
The study was suspended on March 13, 2020, due to restrictions imposed by the COVID-19 pandemic. Enrollment resumed at 4 of the 5 networks on May 26, 2020, and at the 5th network on September 8, 2020. When enrollment resumed, the survey protocol was changed to accommodate COVID-era changes to visits. Staff mailed the study packet after the patient’s visit rather than before due to the large amount of rescheduling of in-person visits to virtual visits that occurred shortly before appointments. The same reminder protocol was followed. Enrollment closed on April 2, 2021.
Patient Survey Outcomes and Measures
Shared Decision-Making Process Scale (primary outcome): this 4-item measure assessed the discussion of (1) stopping screening as an option, (2) reasons to screen, (3) reasons not to screen, and (4) patients’ screening preferences. Individual items were summed to generate a total score (0–4), with higher scores indicating greater shared decision-making. Patients who indicated no discussion received a score of 0. This brief measure has strong evidence of acceptability, feasibility, reliability, and validity.15,16,17 Although there is no clearly established clinical meaningful difference, effect sizes from 0.39SD to 0.88SD are found when comparing sites with and without formal decision support.17
Knowledge: seven multiple-choice knowledge items, adapted from the Colorectal Cancer Screening Decision Quality Instrument, were scored and summed to calculate a total knowledge score (0–100%).18
Patient’s screening preference: a single item assessed patients’ preferred approach to screening with responses: colonoscopy, stool-based test, no screening, and not sure.
Patient’s screening intention: one item assessed how likely the patient was to follow through with their preferred approach on a 5-point scale from Definitely will to Definitely will not.
CRC discussion: one item assessed whether CRC screening was discussed during the visit (yes/no) and if discussed, how much time was spent (<2 min, 2–5 min, > 5 min).
Satisfaction: one item asked “Overall, how satisfied were you with the visit” on a 4-point scale from Extremely satisfied to Not at all satisfied.
Patients also self-reported overall physical and mental health (PROMIS Scale v1.2-Global Health Physical 2a (poor to excellent)19, family history of colorectal cancer, personal history of prior polyp removal, health literacy (Single-Item Literacy Screener)20,21, race, ethnicity, marital status, and education.
Physicians completed a short baseline survey to collect demographics and prior SDM training experience.
The study was powered to detect a small to medium effect size difference in the primary outcome, SDM process score.17 With 500 surveys, assuming an intraclass correlation coefficient of 0.03, the effective patient sample size was estimated at 394, which would enable detection of a difference of 0.28 standard deviations with 80% power and a two-sided significance level of 0.05.
Responders and non-responders were compared to examine potential non-response bias. Patient sample characteristics were compared between arms since randomization occurred at the physician level. Multivariable regression models were used to adjust for potential effects of any unbalanced variables. The hypotheses were evaluated using an intention-to-treat approach and patient outcomes were analyzed based on their physicians’ assigned arm regardless of whether the physician completed the training, received the reminder, or discussed CRC screening.
We tested hypotheses that patients seen by physicians in the Intervention arm would report higher SDM Process scores, be more likely to discuss CRC screening, have greater knowledge, and have higher visit satisfaction than patients of physicians in the Comparator arm. We used a linear regression model with Generalized Estimating Equations (GEE) techniques to account for the patients-within-physicians data structure and compare the continuous outcomes (SDM Process and Knowledge scores) between arms.22 We used logistic regression models with GEE for binary outcomes, ordinal regression for ordered categorial outcomes, and multinomial regression for categorical outcomes (e.g., screening preference). Models included study arm, patient age, patient sex, prior screening (yes/no), physician age, physician gender, physician years since training, and hospital network.
Heterogeneity of Treatment Effects (HTE)
The pre-specified HTE analysis explored the interaction between the study arm and different factors on outcomes. Physician factors included (1) hospital network, (2) gender, (3) age, (4) years in practice, and (5) prior experience with SDM training. Patient factors included (1) sex, (2) age, (3) prior screening history, and (4) overall health. Linear or logistic regression models with the GEE approach were used to test interactions between study arms and these factors. Due to the exploratory nature of the HTE analysis, we reported treatment effects in each subpopulation when the significance level for the interactions between intervention and these factors was ≤0.1.
Patient and Public Involvement in the Research Study
Four patient advisors actively participated on the study team. They attended meetings and provided feedback on study design, training content, communication and messaging to patient participants, and selection of outcomes.