Objectives and hypotheses

The primary objective (Aim 1) is to determine the efficacy of the STOP collaborative care intervention, in comparison to enhanced usual care (EUC), for reducing risky opioid use in adult primary care patients. Risky opioid use is defined as nonmedical use of prescribed opioids (e.g. more than prescribed or non-prescribed opioids) or any use of illicit opioids. We hypothesize that patient participants with PCPs assigned to the STOP intervention, relative to patient participants with PCPs assigned to EUC, will have fewer days of risky opioid use, measured at 6 months post-baseline (primary outcome), and at 3-, 9-, and 12-months post-baseline (secondary outcomes).

Secondary objectives capture patient participant (Aim 2) and provider participant-level outcomes (Aim 3). We hypothesize that patient participants with PCPs assigned to the STOP intervention, in comparison to those with providers assigned to EUC, will have improved measures of other substance use, quality of life, mental health symptoms and no worsening of pain severity. For PCPs in the study (provider participants), we hypothesize that providers assigned to the STOP intervention, in comparison to those assigned to EUC, will adopt improved treatment practices that include safe prescribing of controlled substance and increased monitoring for patients with subthreshold OUD. Specific secondary outcomes are detailed below and in the Additional file 1.

Overview of trial design

The STOP trial is a cluster randomized controlled trial that tests the efficacy of the STOP intervention to reduce opioid use and overdose risk, and to prevent progression of OUD in adult patients with risky opioid use over the course of 12 months, as compared to EUC. It is being conducted within the NIDA Clinical Trials Network, as part of the NIH Helping to End Addiction Long-term (HEAL) Initiative.[29] Five geographically distinct study sites, each having one or more participating primary care clinics, are participating. PCPs are randomized, stratified by site, in a 1:1 ratio to either the STOP intervention or control EUC condition. In the case of one site where multiple PCPs practice as a team (i.e., share patient care among multiple PCPs in the team), the team of PCPs together counts as a single cluster for randomization. Patients who are eligible and enroll receive the intervention according to the assignment of their PCP. EUC consists of primary care treatment as usual, plus receipt of printed educational materials addressing opioid-related overdose prevention and an educational video on cancer screening.

The STOP intervention is a collaborative care model consisting of (1) a practice-embedded nurse care manager (NCM) who provides patient participant education and supports the PCP in engaging and monitoring patients who have risky opioid use; (2) brief advice delivered to patient participants by their PCP and/or prerecorded video message from a study investigator (LG) about health risks of opioid misuse; and (3) at least 2 sessions, and up to 6 sessions, of telephone health coaching to motivate and support behavior change amongst patient participants.

Prior to enrollment, patients are informed that their PCP is participating in a “healthy living study” and are blinded to both the study condition of their PCP and the focus of the study (risky opioid use). After enrollment it is possible that they may be able to deduce their PCP’s assignment, but this information will not be volunteered by clinical or research staff. To assess the success of the blind, patient participants will be asked at their last study visit what they considered to be the goal of the research. Patient participants are assessed through multiple modalities, including monthly brief surveys sent via text message and email, structured questionnaires delivered over the phone and via a web-based portal at baseline and quarterly for 12 months, and urine drug screens at baseline, 6 and 12 months. PCP prescribing and monitoring behaviors are assessed from the electronic health record (EHR) as well as self-administered questionnaires regarding attitudes and behaviors on opioid prescribing. An independent commercial Institutional Review Board (IRB) approved the study, and all sites ceded to this central IRB Fig. 1.

Fig. 1Conceptual model

The STOP collaborative care intervention integrates patient- and provider-facing interventions and is informed by Social Ecological Frameworks for health promotion [30], and the opioid use literature. As depicted in Fig. 2, subthreshold OUD is influenced by multiple factors, including (1) patient-level individual factors such as social determinants of health and mental health disorders; (2) interpersonal factors such as use of opioids by family/friends; (3) clinic organizational factors such as utilization of clinical information systems to enhance communication and efficiency; and (4) community factors such as availability of community resources and other neighborhood characteristics. While STOP does not directly intervene on all the factors that impact a patient’s risky opioid use, it is designed to address each level of the social ecological framework. Additional file 1: Appendix Table S1 maps the STOP intervention components to the Social Ecological Framework’s four levels of influence. Addressing each of these multiple systems of influence is expected to support individuals with subthreshold OUD to reduce their risky opioid use behaviors, which can in turn reduce the risk of escalation to moderate-severe OUD and opioid-related events, such as overdose and death.

Fig. 2

Social Ecological Framework of Risky Opioid Use

Trial registration

Prior to recruitment, the trial was registered at clinicaltrials.gov, identifier NCT04218201.

Site selection

The study team distributed a call for applications to be a clinical site for the trial to the NIDA Clinical Trials Network, which comprises 16 geographically distinct nodes, each of which is affiliated with one or more large academic centers and multiple clinical sites. Eligible sites were expected to have approximately 7–30 PCPs who would be eligible to participate, capacity to provide referrals to opioid treatment program(s), an electronic health record, a clinical champion willing to work with the study team to facilitate study clinical activities, and space for a nurse care manager. Participating sites could not have an existing collaborative care model dedicated to managing patients receiving opioid prescriptions. Although not required, additional desirable characteristics included location in a region with higher-than-average prevalence of opioid use, capacity to provide in-office buprenorphine treatment, ethnic/racial diversity, and space in or near the clinic for research activities.

The study team conducted site visits with potential clinics from June to September 2019 and the sites were finalized in November 2019.

Clinical champion

Each site has a Clinical Champion, (4 sites had one, 1 site had two) who is a physician, or in one case a clinical psychologist, actively practicing in the study clinics. The role of the Clinical Champion is to help recruit PCPs and serve as a resource for the nurse care manager as well as PCPs randomized to the STOP intervention.

PCP eligibility

Eligible PCPs are licensed medical professionals (MD, DO, NP, PA) who have a total weekly patient volume of approximately 40 or more adult patients during a typical week and provide care to approximately 4 or more adult patients receiving chronic opioid therapy and/or with risky opioid use. Chronic opioid therapy is defined as having at least three opioid prescriptions, at least 21 days apart, in the past six months, with EHR documentation of active opioid prescription within 60 days prior to screening. The PCPs must be willing to be randomized to either study condition. They are excluded if they are medical residents or have definite plans to resign from the clinic in the subsequent 24 months or change their schedule in the next 24 months such that they would no longer meet the inclusion criteria for patient volume, per PCP self-report.

PCP recruitment

The Site PI and/or Clinical Champion work with clinic staff to identify potentially eligible PCPs who they approach to assess their interest in participating. Interested PCPs complete a screening survey to assess eligibility and if eligible, research staff conduct informed consent and the PCP completes a baseline survey (see below, study measures). The Clinical Champion delivers orientation to the PCPs randomized to the STOP intervention (see below, training).

Patient eligibility

Patient eligibility criteria include (1) PCP is enrolled in the study. (2) Age ≥ 18 at time of prescreening. (3) Proficient in written and spoken English, as determined by self-report and research staff. (4) Access to phone that can receive text messages. (5) Access to a phone that could receive text messages and to the internet (via smartphone, tablet, computer). (6) Sufficient contact information (minimum of 1 reliable locator). (7) Ability to provide informed consent. (8) Risky opioid use, defined as a Tobacco, Alcohol, Prescription medication, and other Substance Screening TAPS [7] score ≥ 1 for illicit opioid (heroin or fentanyl) use and/or nonmedical prescription opioid use, and/or a positive response (> Never) to Current Opioid Misuse Measure (COMM)[32, 33] items indicating taking more opioid medication than prescribed or opioid medications that belonged to someone else, on any of the following three items: (Item 9) In the PAST 30 DAYS, how often have you needed to take pain medications belonging to someone else?; (Item 14) In the PAST 30 DAYS, how often have you had to take more of your medication than prescribed?; (Item 15) In the PAST 30 DAYS, how often have you borrowed pain medication from someone else?

Exclusion criteria include: (1) Moderate-severe opioid use defined as meeting 4 or more DSM-5 criteria for OUD at screening, as assessed by research staff administering the World Mental Health Composite International Diagnostic Interview (CIDI) modified for DSM-5 criteria (removing legal problems and adding craving) [34]. (2) Receiving medication for OUD or engaged in an opioid treatment program in the past 30 days from screening date. (3) Receiving opioids for end-of-life care. (4) Pregnancy, as determined by self-report. (5) Currently in jail, prison, other overnight facility as required by court of law, or have pending legal action. (6) Definite plans to leave the area or clinic practice within 12 months. (7) Other factors that may cause harm or increase risk to participant.

Patient recruitment

Research staff (research coordinator (RC) or research assistant (RA)) prescreen patients of participating PCPs for eligibility. Although the study was initially conceived as primarily conducting prescreening in clinic waiting rooms, with the COVID pandemic related restrictions, procedures were adapted to offer prescreening in multiple ways, with a reliance on remote methods. Procedures include mailing letters to patients, sending messages to patients through email and/or patient portals, texting and calling patients, posting flyers in clinical areas, and clinical staff referrals. When possible, research staff may also conduct in-person waiting room prescreening.

Patient prescreening for eligibility

The prescreen is conducted via a self-administered web-based “Healthy Living Survey”, which includes questions about general health behavior (exercise, diet) as well as substance use. The purpose of including general health questions is to partially mask the opioid focus of the study for potential participants. For all patients, the prescreening survey includes the TAPS Tool [7]. For patients who report receiving an opioid medication prescription in the previous 6 months, the prescreening tool additionally includes the COMM [32, 33].

To protect anonymity, prescreening forms do not include identifying information for those who are ineligible. If a patient prescreens eligible, they are asked to create a unique ‘study ID’ and leave their contact information for the research staff to reach them. Patients who do not qualify for the study have no further study interaction.

Screening and enrollment

The RC or RA contacts individuals who have prescreened eligible to conduct further ascertainment of eligibility and to administer informed consent. To conduct the eligibility screen, the RC/RA administers the CIDI opioid items and completes a checklist of inclusion and exclusion criteria. The RC/RA obtains written informed consent from eligible and interested individuals.

Individuals who are ineligible on screening due to moderate to severe OUD or receipt of MOUD will receive from the RC/RA information about overdose prevention, a list of community and/or clinic resources for OUD treatment, contact information for a social worker or behavioral health staff member at the clinic, an ineligible patient letter patients can provide to their PCP (for moderate to severe OUD patients only); and the RC/RA will suggest that patients speak with their PCP. They are not informed why they were ineligible.

Study treatment arms

The two study arms include Enhanced Usual Care (EUC) and the STOP intervention. The interventions do not interfere with usual care received by patient participants and could include referral to treatment for OUD if patient participants in either arm of the study should develop moderate-severe OUD during the study. Of note, patients in the study could be receiving prescription opioids from any provider, including the PCP but also non-PCP prescribers.

Enhanced usual care (EUC)

In the EUC arm, PCPs conduct primary care as usual, without the support of the NCM or health coach, and they are not made aware that their patient is enrolled in the trial. The enhancement occurs at the time of the initial baseline visit when patient participants view a short video and receive written materials on cancer screening and educational materials on opioid-related overdose prevention, including how to obtain a naloxone kit. The justification for the EUC is that current practice in primary care generally does not include routine opioid safety measures, including overdose prevention information or naloxone acquisition. The use of the cancer screening video and handouts are to enhance blinding of the study purpose and serve as an attention control.

STOP intervention

The STOP intervention consists of three key components: Brief advice delivered by the PCP and/or a brief video doctor on opioid use, two to six sessions of telephone health coaching, and 12 months of nurse care management. Each component was included in prior clinical trials conducted in primary care settings that were effective in decreasing drug use [16, 17] or for increasing guideline-adherent opioid prescribing [27, 28, 35]. The STOP intervention combines them. Intervention participants also receive educational materials on opioid-related overdose prevention given to EUC participants. See Table 1.

Table 1 Summary of Intervention Provider Roles and Timing of Patient ContactsI. Brief Advice

After a patient participant enrolls, the RA/RC informs the patient’s PCP and provides the PCP with a printed ‘Summary to the Clinician’ that explains the patient participant’s prescreening results (TAPS and, if applicable, COMM) and a script that guides the PCP in providing brief advice on reducing risky opioid use. The scripts are adapted from those used in the QUIT trial [16, 17] and are available in two versions—one for patients misusing illicit opioids and the other for patients misusing prescribed opioids. See Additional file for copies of scripts.

For patient participants with an upcoming scheduled primary care visit at the time of enrollment, PCPs are asked to give brief advice at the time of the visit. If the participant does not have a scheduled primary care visit, the PCPs are asked to call the patient to deliver the brief advice within 10 business days. Afterwards, PCPs fill out an intervention checklist about the advice given (or not given).

All patient participants are asked by research staff to watch a ‘Video Doctor’ recording during the baseline study visit to supplement any brief advice that is delivered by the PCP, as utilized in the QUIT trials [16, 17]. The 5-min video features a 3-min clip from a study investigator (LG) delivering the same brief advice script that is given to the PCP, as well as videos of the health coaches and local nurse care manager introducing themselves and their roles.

II. Telephone health coach

The Telephone Health Coaches (hereafter referred to as Coaches) deliver a series of individual health coaching sessions. The coaches are not affiliated with the clinics but provide a national phone bank available to patients of any of the study clinics. This allows patients to discuss clinic related barriers they face and to discuss issues that they do not want recorded in their EHR. The Coaches utilize elements of Motivational Interviewing (MI) and Cognitive Behavioral Therapy (CBT) to improve patient participant motivation and self-efficacy to change health behavior and behavioral outcomes [36,37,38]. Health coaches may be social workers, counselors, or individuals with other professional training that qualifies them to deliver the STOP counseling intervention.

All patient participants in the STOP arm are offered the two initial health coaching sessions with or without video, based on patient participant preference, at approximately 2- and 4-weeks post-baseline. See Table 1 for content of coaching sessions.

Enhanced coaching (Step 2):

Patient participants who may benefit from additional coaching following the Step 1 sessions are offered up to 4 additional sessions with the same Coach. These sessions ideally start within two weeks of the patient participant’s referral into Step 2 and occur once a week for four consecutive weeks. Enhanced coaching may be recommended by the NCM or PCP, or offered at the discretion of the Coach, based on patient preference as well as clinical criteria which may include ongoing or escalating nonmedical or illicit opioid use, overdose risk behavior, symptoms of OUD, overdose episodes, and/or consequences of opioid and other substance use. Enhanced coaching is offered by the Coach at the time of the second Step 1 session, and patients are free to decline. (Table 1).

III. Nurse care manager

The Nurse Care Manager (NCM) works closely with the intervention PCP to assess, educate, and manage patient participants with risky opioid use. Details of the NCM intervention are outlined in Table 1. Patient participants are asked to schedule an initial visit with the NCM as soon as possible following study enrollment. The NCM continues working with patients in the STOP condition throughout their 12 months of study participation. Following the initial visit with the NCM, the frequency of visits depends on patient participant needs. Throughout the health coaching period, the Coaches and NCMs meet weekly to discuss their patients and strategize how to effectively work with the entire care team to support the patient.

Training

Each of the professionals involved in the intervention receives specific training prior to patient recruitment.

The Clinical Champions receive training conducted by two primary care physicians with expertise substance use (LG and Lead Investigator JM). The primary goal of the training is to prepare Champions to educate and provide guidance to their PCP colleagues. The training orients them to the study and covers the Summary to the Clinician, delivery of brief advice, management and monitoring of patients receiving prescription opioids, identifying, and assessing OUD, treatment of OUD, and overdose prevention.

PCP participants randomized to the intervention receive a 20–30-min training delivered by the site Clinical Champion. This training includes practicing delivery of brief advice, understanding the Summary to the Clinician, documentation on the intervention checklist, and collaboration with the NCM. PCP participants in the EUC condition only receive an introduction to the study during the Clinical Champion’s initial presentation to the clinic.

Health Coaches undergo initial training and then participate in a weekly virtual Learning Community (as in the QUIT trials [16, 17]) for ongoing technical support, mutual problem solving, supervision, and to maintain fidelity of the intervention. Some of the training sessions occur jointly and some are role specific. Study trainers include a PhD clinical psychologist (TL) and a physician expert in health coaching for risky substance use (LG). Training is delivered virtually over a 6-day period and is focused on the content and procedures involved in behavior change coaching, with special focus on building competence in motivational interviewing and cognitive behavioral therapy techniques. Coaches and NCMs learn techniques to refer patient participants to local resources as well as to assess suicidality. After formal training, the health coaches conduct mock sessions, which are audio recorded and coded for fidelity by the trainers.

NCMs undergo 10 h of didactic training over the course of 5 days, and an additional 4–8 h of interactive workshop training prior to the start of their participation in the study. These sessions are led by two experienced nurse care managers and the study Lead Investigator (DB, CA, JL), all of whom participated in prior trials using NCM for opioid prescribing. Throughout the study, these trainers facilitate a weekly hour-long group meeting of the NCMs to offer technical support.

PCP participant compensation

Provider participants receive compensation in the form of debit cards or gift cards for completing surveys ($25 per survey, total of $50).

Patient participant compensation

Patient participants receive compensation in the form of debit cards or gift cards for completing study activities including surveys, urine drug screening, and updating contact information (total possible $670 for full participation). Intervention patients also get $3 per telephone health coaching to offset the cost of data usage on mobile phones (maximum $18) but are not otherwise compensated for participation in any aspect of the intervention. A subsample of intervention patients who undergo qualitative interviews to assess the intervention experience receive a further $50.

Study assessmentsSources of data

Assessments and times of their administration are listed in Table 2. Outcome data are collected from patient participants via self-administered surveys, RA/RC-administered surveys, and urine toxicology tests. Self-administered assessments were selected whenever possible, to increase feasibility and limit social desirability bias. Data from PCP participants are collected via self-administered surveys, and through audit of the electronic health record. In addition, a research staff member conducts qualitative interviews by telephone with a convenience sample of patient participants between three and six months of study participation, with distribution across sites.

Table 2 Aims, Outcomes and Measures

Patient participants are administered the “Healthy Living Survey” at baseline and monthly via email/text with a computerized link to questions quantifying use of opioids and non-opioid substances, and prescription opioid misuse. Questions on diet, exercise, and stress are included to blind the study participants as to the study focus.

Primary Outcome: Days of Risky Opioid Use.

The primary outcome measure is self-reported number of days of risky opioid use in the past 180 days, assessed at 6 months after the baseline visit using single item questions based on the Addiction Severity Index [39, 40] and the COMM [32, 33] via the “Healthy Living Survey” (See Text box below). Participants are asked to specify the number of days of illicit opioid use and of nonmedical use of prescribed opioids in the past 30 days (range is 0–30 days). Illicit opioid use includes use of heroin or fentanyl. Nonmedical opioid use includes using prescribed opioids more frequently or at higher doses than instructed on the prescription (e.g., taking 2 tablets when the prescription indicates a dose of 1 tablet), or taking pharmaceutical opioids that were not prescribed to them. The extent of overlap between these different types of risky opioid use is also measured, when applicable. The measure is calculated as the sum of all days of use reported on the assessments of past 30-day drug use for the first 6 months (i.e., the sum of days of use from the measures collected on day 30, day 60, day 90, day 120, day 150, and day 180 minus the overlap). For example, if a participant reports nonmedical opioid use on 10 days, illicit opioid use on 12 days, and there were 4 days on which they used both nonmedical and illicit opioids, then the total number of days of risky use are calculated as 10 + 12–4 = 18.Days of risky opioid use are assessed at the prescreen eligibility survey and every 30 days for the 12 months of study participation. The assessment is self-administered using a computerized form delivered using a text message or email

link to a monthly Healthy Living survey.

Secondary outcomes

Substance Use: The Healthy Living Survey includes other items on opioid and other substance use for secondary analyses. Substances queried using additional questions based on the Addiction Severity Index [39, 40] include the number of alcohol heavy drinking days (4 + drinks/day for women, 5 + drinks/day for men), and days of nonmedical use of benzodiazepines or prescription stimulants,, illicit stimulant (cocaine or methamphetamine), marijuana, and other drugs. Additional items adapted from the COMM [32, 33] include: “During the PAST 30 DAYS, on how many days did you borrow pain medication from someone else?” and “During the PAST 30 DAYS, on how many days did you use your pain medication for symptoms other than for pain (e.g., to help you sleep, improve your mood, or relieve stress)?”. Days of non-opioid substance use is measured using the same approach as days of risky opioid use, with one question asking about number of days of use of each substance type in the past 30 days, but without questions about overlapping days of use.

Other sources of data on substance use include qualitative urine drug screens to verify self-reported drug use at baseline, 6, and 12 months. The urine drug screen measures opioids, oxycodone, fentanyl, barbiturates, benzodiazepines, cocaine, amphetamine, methamphetamine, marijuana, methadone, buprenorphine, phencyclidine and ecstasy. Additionally, the RC/RA administers an assessment of risky opioid use via Timeline Follow-Back method at 90 and 180 days. This assessment will measure days of risky opioid use should the monthly assessments yield low response rates. The Psychiatric Diagnostic Screening Questionnaire (PDSQ) [41], is a brief self-administered instrument to assess DSM-IV alcohol and drug use disorders. The PDSQ was chosen as the only validated self-administered tool for DUD and AUD and was validated in more than 3000 patients in both medical and mental health outpatient settings [39, 42, 43]. Because it does not distinguish which substance meets criteria for DUD, the CIDI is administered by the RA/RC for OUD at baseline, 6, and 12 months. Overdose risk is assessed by the Overdose Risk Behavior Questionnaire, [15, 44,45,46] and past month overdose experience is assessed through a single question used in prior research.[47].

For non-substance use measures, see Table 2.

Primary care provider outcomes focused on improving guideline-concordant practices with respect to patients with risky opioid use, including prescribing lower doses of opioids, avoiding concomitant benzodiazepines, increased monitoring and higher visit frequency. Provider-level outcomes are measured through chart abstraction in the Electronic Health Record, including opioid prescriptions (< 50 Morphine Milligram Equivalent (MME), 50–90 MME and =  > 90 MME), benzodiazepine prescriptions, naloxone prescriptions, number of urine drug screens ordered and completed, new diagnoses of OUD, and number of scheduled primary care visits per patient participant.

Other outcomes

Other measures will be collected to assess exploratory outcomes, to characterize domains from our conceptual model, and may be used to adjust models of primary and secondary outcomes. See Table 2 for details.

Qualitative interviews of up to 30 participants in the intervention condition after 3 months will explore how they engaged with the different components of the intervention (NCM, Health Coach, PCP brief advice). Qualitative interviews are guided by the social ecological framework [48] to examine barriers and facilitators to reducing opioid use and intervention engagement related to patient characteristics, social networks, the community, and the health care system. Interviews will be audio-recorded, transcribed, and analyzed using thematic analysis techniques [49].

Study assessments are estimated to require one to two hours at baseline and 3, 6,9 and 12 months. The Monthly health living survey takes participants approximately 15 min.

Fidelity measures

To evaluate fidelity to the STOP intervention model, each person (PCP, NCM, Health Coach) completes measures about interventions delivered. For the PCP, this includes a checklist reporting what elements of the brief advice they delivered, and the time required. The NCMs document delivery of the intervention including number of interactions, type of interactions (e.g., text, phone, in-person visit), and the duration and content of those interactions. Health coaches complete and session content checklist at the conclusion of each session. In addition all coaching calls are recorded (when patient participant gives permission) and coded by a trained, independent rater for use of motivational interviewing and cognitive behavioral therapy, using the Motivational Interviewing Treatment Integrity (MITI) Scale 4.1 [50, 51] and following the strategy of Haddock and colleagues to integrate elements of CBT into the fidelity monitoring [51]. All sessions are recorded for the first three months of the study to verify that Health Coaches are achieving consistent delivery of the intervention. Afterwards, a random 10% of the sessions are evaluated for adherence to the intervention.

At the end of study participation, the last survey that all patient participants complete is an assessment of what they think the study was targeting (opioids, diet, exercise, etc.).

Analysis

A mixed effects negative binomial model with a log link will be fit to estimate the difference in mean number of days of risky opioid use in the past 6 months between treatment and control groups. The model will include random PCP intercepts to account for within-PCP correlation of participant response values, fixed site effects to account for within-site correlation of responses, a fixed treatment effect, and the baseline value of the response variable (days of use within 30 days prior to the baseline assessment timepoint), which may improve precision. Letting Yjk denote the response variable of a participant in site j with PCP k, treatment denote a binary treatment indicator variable, and baseline denote the baseline value of the outcome variable, the formula for the log-transformed expectation of Yjk is below:

$$}\left[ }\left( }_}}} } \right)} \right] = \beta _ + \beta _ \cdots } + \beta _ \cdots } + \alpha _}} + \gamma _}}$$

where αj is the fixed effect for site j and γk is the random intercept for PCP k. We assume the PCP random intercepts are normally distributed with mean zero and standard deviation σγ. The negative binomial model allows for potential overdispersion of the response variable and is more flexible than a Poisson distribution which would require the mean to be equal to the variance. If 2 or more PCPs are randomized as a single cluster because they practice as a team, they will be analyzed as a single cluster. The primary hypothesis will be evaluated by testing whether the treatment effect, β1, is different from zero. This is equivalent to testing whether the control mean is different from the intervention mean.

The primary analysis will be performed on the Intention-to-Treat (ITT) principle, analyzing patient participants according to their randomization assignment regardless of potential exposure to the opposite assignment. The primary analysis will use multiple imputations to account for missingness of the primary outcome variable. The imputation model will incorporate information on days of risky use from the timeline follow back surveys. Further details of the imputation model will be specified in the Statistical Analysis Plan, which will be finalized prior to the primary analysis is performed. A secondary analysis will be performed on the Per Protocol (PP) population. This population will exclude patient participants who had appointments with PCPs in the opposite treatment arm. The primary outcome will be evaluated using a two-sided test with a type I error rate of 5%.

Secondary analyses of the primary outcome include the model used in the primary analysis adjusted for individual-level covariates which may be associated with the response.

Analyses of secondary outcomes

A secondary analysis will fit the same model to the 180-day outcome measured at the 12-month timepoint to assess durability. Further secondary analyses will explore temporal patterns of number of days of risky opioid use. In addition, a mixed effects negative binomial model will be fit including treatment by time interactions for each timepoint. The model will include participant random intercepts to account for correlation in responses arising from multiple observations for each patient participant.

Analyses for Aim 2 secondary outcomes will be analyzed with mixed effects regression models with treatment by time interactions (analogous to those described above for the primary outcome) or generalized linear models with generalized estimating equations (GEE) to account for correlation of responses will be fit. Similar regression models will be fit for the Aim 3 objectives, but these will use provider participant rather than patient participant as the unit of observation.

Sample size estimation/Statistical Power

Power was calculated based on simulations exploring four possible scenarios to consider different time trends of the intervention effect as well as a possible assessment impact in the control group. Assessment impact means that the regular recording of opioid use in the monthly assessments causes self-awareness which leads to behavior modifications [52, 53]. The four scenarios are depicted in Additional file