Collaborative care intervention for risky opioid use among primary care patients: The STOP randomized clinical trial Authors: Jennifer McNeely, Geetha A Subramaniam, Rebecca Stone, Shayna Mazel, Noa Appleton, Yasna Rostam‐Abadi, Amanda M Bunting, Yidan Shi, Travis I Lovejoy, Lillian Gelberg, Donna Beers, Margaret Kline, Song Zhang, Tobie Kim, Ashley Case, Jennifer McCormack, Michael A Incze, Jane M Liebschutz, the STOP trial site principal investigators Published: 2026 ABSTRACT AbstractBackground and aimsIndividuals who engage in illicit or nonmedical opioid use may have elevated risk of health and social consequences, including progression to opioid use disorder (OUD). Preventive interventions to reduce this risk are lacking. This trial tested the impact of a primary care‐integrated collaborative care approach for reducing risky opioid use, defined as nonmedical use of prescription opioids or any use of illicit opioids.DesignCluster‐randomized controlled trial randomized primary care providers (PCPs) and their patients into the Subthreshold Opioid Use Disorder Prevention (STOP) intervention or enhanced usual care (EUC).SettingPrimary care clinics at 5 U.S. sites.ParticipantsPCPs and their patients were recruited January 2021–May 2023. A total of 119 PCP clusters (STOP = 48, EUC = 51) and 202 patients (STOP = 88, EUC = 114) enrolled. Eligible patients were adults (≥18 years) having current risky opioid use, without moderate–severe OUD. Patient participants were majority female (63.4%), white (70.8%) and non‐Hispanic (96.5%), with a mean age of 55.7 [standard deviation (SD) = 12.7] years. At baseline, 63.4% of participants had moderate–severe pain (Brief Pain Inventory) and below average physical (79.2%) and mental (62.4%) health (SF‐12).InterventionsThe STOP collaborative care intervention consisted of brief advice from the PCP about reducing risky opioid use, meetings with a clinic‐embedded nurse care manager over 12 months and remote health coaching (2–6 sessions). Both groups received primary care treatment as usual and overdose risk reduction materials.MeasurementsThe primary outcome was total days of risky opioid use, recorded from 6 monthly electronic surveys. A key secondary outcome was moderate–severe OUD at 6 and 12 months.FindingsA total of 77 (87.5%) STOP and 107 (93.9%) EUC participants completed the 6‐month assessment period. The primary outcome analysis used the Intention‐to‐Treat sample with multiple imputations of missing data. Mean days of risky opioid use at 180 days were lower in STOP than EUC [12.2 (SD = 27.73) vs. 15.5 (SD = 32.64)]; the difference between groups adjusted for baseline risky opioid use was not statistically significant (rate ratio 0.95, 95% confidence interval = 0.52–1.74). One STOP participant (1.1%) and 13 EUC participants (11.4%) developed moderate–severe OUD at 6 months, and 3 (3.4%) STOP and 6 (5.3%) EUC participants had moderate–severe OUD at 12 months (P < 0.001).ConclusionsThis cluster‐randomized controlled trial did not find evidence that the STOP intervention for reducing risky opioid use produced greater reductions over 6 months compared with enhanced usual care, though fewer intervention participants progressed to moderate–severe opioid use disorder. Patients had a high burden of pain and comorbidities that may present challenges to reducing opioid use.Keywords: collaborative care, opioid misuse, opioid‐related disorders, pain, prevention, primary care, substance use disorder INTRODUCTION Evidence‐based interventions are lacking for the nearly 9 million Americans who engage in illicit or non‐medical opioid use but do not meet the diagnostic criteria for moderate or severe opioid use disorder (OUD) [1, 2]. While medications for OUD are the first‐line treatment for individuals with moderate–severe OUD—defined by meeting four or more of the criteria listed in the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM‐5) [3]—effective approaches to promote safety and prevent harms are lacking for those who fall below this threshold [4]. Even low levels of illicit opioid use or the non‐medical use of prescribed opioids may place individuals at risk of developing moderate–severe OUD, overdose or other opioid‐related health and social consequences [5, 6, 7, 8]. We refer to this opioid use behavior as ‘risky opioid use’ [9, 10], though it may also be called opioid misuse or unhealthy opioid use. The prevalence of risky opioid use is approximately 5% among the general population of primary care patients [11, 12], and is higher (21%–29%) among patients prescribed chronic opioid therapy [13]. Individuals with risky opioid use may experience few opioid‐related symptoms and rarely seek addiction treatment [13, 14]. However, primary care providers (PCPs) regularly encounter these patients during routine medical care, making primary care a touchpoint for mitigating risky opioid use. We undertook the Subthreshold Opioid Use Disorder Prevention (STOP) trial to test a primary care–integrated collaborative care approach to reducing risky opioid use among adult primary care patients. Collaborative care has been shown to improve outcomes for depression [15, 16, 17, 18], other mental health conditions [19, 20] and chronic pain, yet the evidence supporting the use of this model to address substance use is limited. Prior studies of collaborative care for substance use in the USA have enrolled patients diagnosed with alcohol or opioid use disorders, rather than those with earlier stage risky opioid use, and have had mixed results [21, 22, 23, 24, 25, 26]. Outside the USA, parallel models have incorporated task‐sharing interventions whereby non‐specialist health workers delivered behavioral interventions, referrals and care management [27, 28, 29] to patients with substance use disorders, although none have evaluated models for risky opioid use. Collaborative care models typically consist of a PCP working with a care manager who maintains a clinical registry to proactively track and manage care delivery [30]. The STOP intervention was an adapted collaborative care model incorporating PCP brief advice, an in‐clinic nurse care manager (NCM) and telephone health coaching [9]. The primary objective was to test the impact of the intervention on the number of days of risky opioid use over a period of 6 months. A secondary objective was to explore the impact of the intervention on preventing progression to moderate or severe OUD. METHODS Study design A cluster randomized clinical trial, randomized at the PCP level, tested the efficacy of the STOP intervention compared with enhanced usual care (EUC) for reducing risky opioid use among adult primary care patients (see published protocol) [9]. Data collection occurred between January 2021 and June 2024, during which time enrolled patients completed self‐administered assessments over 12 months of study participation. Five study sites participated, each having one or more participating primary care clinics. Three sites were in academic medical centers, one was a federally qualified health center and one was a rural primary care practice. Site locations and characteristics are presented in Table S1. This report follows the Consolidated Standards of Reporting Trials (CONSORT) guideline [31]. The study was approved by the single Institutional Review Board of the Biomedical Research Alliance of New York and was registered with ClinicalTrials.gov (NCT04218201). Participants The eligibility requirements for PCPs included [9]: (i) licensed physician, nurse practitioner or physician assistant; (ii) an average weekly patient volume of ≥40 adult patients; and (iii) four or more adult patients with risky opioid use and/or receiving chronic opioid therapy. Eligible patients [9] had risky opioid use, defined as illicit opioid use and/or non‐medical prescription opioid use in the past 3 months measured on the Tobacco, Alcohol, Prescription medications, and other Substances (TAPS) Tool, and/or non‐medical use of prescription opioids (specifically, taking more opioid medication than prescribed or taking opioid medications belonging to someone else) as reported on the Current Opioid Misuse Measure (COMM) [32]. Other eligibility criteria were: (i) having an enrolled PCP; (ii) age ≥18 years; (iii) proficiency in written and spoken English; and (iv) access to device(s) with text messaging and internet capability. Exclusion criteria were: (i) moderate–severe OUD, as assessed by the World Mental Health Composite International Diagnostic Interview (CIDI) modified for DSM‐5 criteria (removing legal problems and adding craving) [33, 34, 35]; (ii) receiving medication for OUD in the past 30 days; (iii) receiving end‐of‐life care; (iv) pregnancy; (v) current prisoner or pending legal action; and (vi) definite plans to leave the clinic within 12 months. Procedures Recruitment and consent Recruitment occurred between January 2021 and May 2023. Research staff invited PCPs practicing in the study clinics to screen for eligibility, and eligible PCPs who agreed to enroll provided written informed consent. Subsequently, research staff invited patients of enrolled PCPs to complete an electronic self‐administered pre‐screening questionnaire. Although the study had planned to conduct pre‐screening in clinic waiting rooms, procedures were adapted in response to COVID‐related restrictions to offer pre‐screening in multiple ways, with a reliance on remote methods (electronic health record messages, phone calls, mailed letters) [9]. Research staff attempted to contact all adult patients of enrolled PCPs, and often conducted more frequent outreach to patients with substance use‐ or mental health‐related diagnoses. To reduce discomfort and encourage the accurate disclosure of substance use, pre‐screening questionnaires were self‐administered, anonymous and included questions about general health behavior (exercise, diet) as well as substance use. Patients who pre‐screened as eligible could either contact the research staff themselves or leave their contact information. Because of concerns that patients would feel stigmatized or be unwilling to report substance on screening and assessment surveys, the study was called the ‘Healthy Living Study’ and participants were not informed of the focus on opioid use. A similar approach was used in the Quit Using Drugs Intervention Trial (QUIT) early intervention trial, which informed the design of our study [9, 10]. Research staff confirmed eligibility and administered written informed consent. Randomization The PCP participants were randomized, stratified by site, in a 1 : 1 ratio to the STOP or EUC condition. A single PCP defined a cluster, except at one site where multiple PCPs practiced in teams and each team was randomized as a single cluster. Enrolled patients received the intervention to which their PCP was assigned. Assessments Patient assessments were both self‐administered electronically and administered by research staff. Self‐reported data were collected from patient participants for 12 months via two methods: (i) brief monthly surveys sent via text message and email; and (ii) structured questionnaires completed at baseline and quarterly, with components delivered over the phone by research staff and self‐completed via a web‐based portal. Assessments included items about general health behaviors, to lessen the focus on opioids. To assess blinding, participants were asked during their final study visit what they considered to be the goal of the research. Baseline assessments collected demographic and clinical characteristics, including chronic pain symptoms (Brief Pain Inventory) [36] and the 12‐item Medical Outcomes Study Short Form (SF‐12) [37]. Monthly surveys collected self‐reported number of days of use in the past 30 days of illicit opioids, non‐medical use of prescribed opioids and non‐opioid substance use, using questions based on the Addiction Severity Index [9]. Additional self‐reported assessments of substance use collected at baseline, and again at 6 and 12 months, were: (i) adapted CIDI modified for DSM‐5, to assess for moderate–severe OUD [35]; (2) COMM, administered to participants who self‐reported having prescribed opioid medications [32]; (iii) Overdose Risk Behavior Questionnaire [39, 40, 41, 53] and a single question about overdose experiences in the past 6 months [42]; and (iv) readiness to change substance use (not assessed at 12 months) [43]. Interventions All participants received primary care treatment following usual practices in their clinics, without limitations related to the study. STOP intervention group Participants in the STOP intervention received a modified collaborative care model with three key components: PCP brief advice; an NCM; and health coaching. Site clinical champions trained PCPs to give brief advice (approx. 3 minutes) using a script adapted from the QUIT trial [10] and based on the Brief Negotiated Interview that asked permission to discuss risky opioid use, gave feedback and education about opioid‐related risks, and recommended reducing use. PCPs delivered brief advice during visits or via phone calls within 10 business days of enrollment. Participants also viewed a video of a clinician providing brief advice about reducing risky opioid use. The PCP scripts and videos differed depending on whether the patient reported non‐medical prescription opioid use, illicit opioid use or both. Site clinical champions trained PCPs to give brief advice (approx. 3 minutes) using a script adapted from the QUIT trial [10] and based on the Brief Negotiated Interview that asked permission to discuss risky opioid use, gave feedback and education about opioid‐related risks, and recommended reducing use. PCPs delivered brief advice during visits or via phone calls within 10 business days of enrollment. Participants also viewed a video of a clinician providing brief advice about reducing risky opioid use. The PCP scripts and videos differed depending on whether the patient reported non‐medical prescription opioid use, illicit opioid use or both. The clinic‐based NCM worked with participants and intervention PCPs throughout the 12 months of study participation. NCM activities included overall care management, health education, counseling on risk reduction and overdose prevention, and facilitating linkage to specialty care, including for treatment of pain, mental health and addiction, as indicated. The clinic‐based NCM worked with participants and intervention PCPs throughout the 12 months of study participation. NCM activities included overall care management, health education, counseling on risk reduction and overdose prevention, and facilitating linkage to specialty care, including for treatment of pain, mental health and addiction, as indicated. Telephone health coaching [10, 44], delivered by paraprofessionals (master’s‐trained professionals without clinical licensure) via phone or videoconference, based on participant preference, utilized elements of motivational interviewing (MI) and cognitive behavioral therapy (CBT) to improve patient participant motivation, self‐efficacy and skills to change health behavior, including risky opioid use [45, 46, 47]. All participants were asked to complete two 40‐minute health coaching sessions during the first month, which were used to establish rapport and enhance motivation to change risky opioid use. Participants could then receive up to four additional sessions that used CBT to address risky opioid use and other health‐related concerns (e.g. chronic pain, insomnia, depression, stress). The health coaches and NCMs met weekly to discuss shared patients. Telephone health coaching [10, 44], delivered by paraprofessionals (master’s‐trained professionals without clinical licensure) via phone or videoconference, based on participant preference, utilized elements of motivational interviewing (MI) and cognitive behavioral therapy (CBT) to improve patient participant motivation, self‐efficacy and skills to change health behavior, including risky opioid use [45, 46, 47]. All participants were asked to complete two 40‐minute health coaching sessions during the first month, which were used to establish rapport and enhance motivation to change risky opioid use. Participants could then receive up to four additional sessions that used CBT to address risky opioid use and other health‐related concerns (e.g. chronic pain, insomnia, depression, stress). The health coaches and NCMs met weekly to discuss shared patients. Enhanced usual care (EUC) group The EUC participants watched a brief video of a non‐study clinician providing advice on cancer screening that served as an attention control and received educational materials on opioid‐related overdose prevention, including how to obtain naloxone. PCPs in the EUC condition were not notified about trial‐enrolled participants. Measurements Primary outcome The primary outcome, defined as the sum of all days of risky opioid use for 180 days following baseline, was assessed using single‐item questions based on items from the Addiction Severity Index [48, 49] and the COMM [50, 51] (see published protocol [9]). On monthly self‐administered surveys, participants reported days of illicit opioid use and days of non‐medical use of prescribed opioids in the past 30 days, and the primary outcome measure was calculated as the sum of all days of use for the first 180 days of study participation. The use of both illicit and non‐medical opioids on the same day counted as one day of use. Secondary outcomes The severity of OUD (DSM‐5 criteria) [35] was measured using the CIDI assessment administered and categorized as none or mild OUD (0–3 criteria), or moderate–severe OUD (4+ criteria). Days of risky opioid use over months 7–12 (days 181–360 post‐baseline) were measured as a secondary outcome. Prescription opioid risk behavior was measured using the COMM, with a cut‐off of ≥9 indicating high risk [50]. Non‐opioid substance use measures collected at baseline, 6 months (days 1–180) and 12 months (days 181–360) were days of heavy alcohol use (4+ drinks/day for women, 5+ drinks/day for men), and the non‐medical use of benzodiazepines, prescription stimulants, illicit stimulants (cocaine or methamphetamine), cannabis and other drugs. The Overdose Risk Behavior Questionnaire was used to calculate an overdose risk score (range = 0–34) [39, 40, 41, 53]. Self‐reported readiness to change substance use (alcohol, opioid or other substance use) was measured on a scale of 1–10, with 10 representing the highest readiness [43]. Statistical analysis Sample size Power simulations considered different time trends of the intervention effect as well as possible assessment impacts in the control group. We based power calculations on previous trial data [9], with mean days of risky opioid use within the past 90 days estimated as 20 days at baseline, and with differences between control and treatment groups of 10–12 days (within past 180 days) at follow‐up. We drew numbers of days of risky opioid use from binomial distributions for the two 90‐day time intervals, and selected variance parameters to generate data, with most values being less than 30 and with a few values of over 45 days (per 90 days). Assuming 20% loss to follow‐up, we originally estimated that enrolling 60 PCPs and eight participants per PCP for a total of 480 participants would provide 95%–99% power for a two‐sided significance level of P ≤ 0.05. In November 2021, recruitment challenges led to a re‐evaluation of sample size; additional simulations indicated that with 20% loss to follow‐up, power was 89% or greater with enrollment as low as 50 PCPs and 200 participants. The trial was not powered for any secondary outcomes. Primary outcome The primary outcome analysis used the intention‐to‐treat sample with multiple imputation of missing primary outcome data. Missing monthly survey data were imputed assuming missing at random (MAR) and missing not at random (MNAR) for intermittent and dropout survey missingness, respectively. Following standard interpretation, intermittent missingness was assumed to be MAR because the participants remained in the study while only missing occasional monthly surveys; in this case, missing values were imputed (predicted) using observed values. Participants who dropped out from the study (i.e. missed all subsequent surveys after their last completed survey), were assumed to have MNAR missingness, which is not predictable using observed values. Instead, days of use after study dropout was assumed to be higher in the STOP group and was imputed as resembling that of the EUC group (control‐based imputation) [52]. The protocol‐specified primary outcome analysis used a mixed‐effects negative binomial model with a log link to estimate the mean difference in days of risky opioid use between STOP and EUC groups, with fixed effects for treatment, site and the baseline days of risky opioid use, and random PCP intercepts to account for within‐PCP correlation of participant responses. However, the random PCP intercept did not improve estimation (likelihood ratio test, P ≈ 1) and was dropped, resulting in a final model with fixed effects only. The rate ratio, defined as the exponentiated estimate of the treatment effect, is the ratio of mean total days of risky opioid use in STOP compared with EUC. Appendix S1 presents the results of pre‐specified supplemental analyses of the primary outcome. Given the high number of zero opioid use days in the sample, these analyses included a zero‐inflated negative binomial model. Secondary outcomes Secondary outcomes were examined for the EUC and STOP groups using descriptive analyses. Frequencies of dichotomous and multi‐level categorical variables were compared using chi‐square tests. Continuous variables were compared using Student’s t‐tests or Wilcoxon rank‐sum tests, as appropriate, based on distributional assumptions. Proportions of moderate–severe OUD were compared between groups at 6 and 12 months, to test the hypothesis that participants in the STOP group would have a lower incidence of moderate–severe OUD. Group differences in substance use were assessed for each substance class at baseline (30 days prior to enrollment), 6 months (days 1–180) and 12 months (days 181–360). Comparisons were conducted using both a binary indicator of any use and the number of days of use. Additional group comparisons were: (i) high‐risk prescription opioid use behavior, measured in participants with an opioid prescription; (ii) overdose risk behavior; (iii) self‐reported non‐fatal overdose events; and (iv) readiness to change substance use. All analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC, USA). RESULTS Screening and enrollment Study staff screened 157 PCPs and enrolled and randomized 119 of them (86.2% of the eligible PCPs) (Figure S1). Patients were enrolled into 33 PCP clusters in EUC and 31 PCP clusters in STOP. A total of 42 589 pre‐screens were completed by patients, resulting in 686 eligible pre‐screens of which 522 were successfully contacted for screening (Figure 1). Among the 242 patients who screened eligible, 202 (83.5%) enrolled and were included in the ITT analysis (114 in EUC, 88 in STOP). A total of 104 (91.2% participants in the EUC group and 76 (86.3%) in the STOP group completed the study. There were no significant differences in baseline demographic characteristics between completers and non‐completers. Complete primary outcome data (i.e. no missing data on days of risky opioid use collected on baseline or monthly surveys) was available for 90 (78.9%) EUC participants and for 65 (73.9%) STOP participants. Flow diagram of patient participant enrollment. Participant characteristics The baseline demographic characteristics of participants were similar between treatment groups, excepting sex (Table 1). The mean age was 55.7 years (SD = 12.7 years), and the majority was female (55.3% in EUC, 73.9% in STOP) and White (70.8%). A higher proportion in the STOP group reported fair or poor general health (56.2% in EUC, 65.9% in STOP). TABLE 1Demographic characteristics of participants by treatment condition, at baseline.OverallEUCSTOP P a n = 202 n = 114 n = 88 Demographic characteristics Age, years 0.12Median (range)58.0 (18.0–79.0)58.0 (23.0–79.0)58.0 (18.0–75.0)Mean (SD)55.7 (12.7)56.9 (12.0)54.09 (13.3) Sex, n (%) 0.01 Female128 (63.4)63 (55.3)65 (73.9)Male74 (36.6)51 (44.7)23 (26.1)Missing n 0 (0.0)0 (0.0)0 (0.0) Hispanic or Latinx, n (%) 0.67No195 (96.5)109 (95.6)86 (97.7)Yes7 (3.5)5 (4.4)2 (2.3)Missing n 0 (0.0)0 (0.0)0 (0.0) Race, n (%) 0.65White143 (70.8)83 (72.8)60 (68.2)Black or African American43 (21.3)22 (19.3)21 (23.9)Other b 14 (7.0)7 (6.1)7 (8.0)Do not know/refused to answer2 (1.0)2 (1.8)0 (0.0)Missing n 0 (0.0)0 (0.0)0 (0.0) Education, n (%) 0.20Less than high school diploma12 (5.9)7 (6.1)5 (5.7)High school graduate or GED or equivalent41 (20.3)27 (23.7)14 (15.9)Associate’s degree or some college, no degree71 (35.2)34 (29.8)37 (42.1)Bachelor’s degree41 (20.3)21 (18.4)20 (22.7)Postgraduate or professional school degree37 (18.3)25 (21.9)12 (13.6)Missing n 0 (0.0)0 (0.0)0 (0.0) Employment, n (%) 0.15Working now68 (33.7)33 (29.0)35 (39.8)Retired62 (30.7)41 (36.0)21 (23.9)Disabled39 (19.3)24 (21.1)15 (17.0)Unemployed15 (7.4)5 (4.4)10 (11.4)Other c 18 (8.9)11 (9.6)7 (8.0)Missing n 0 (0.0)0 (0.0)0 (0.0) Insurance coverage, n (%) 0.34Private85 (42.1)47 (41.2)38 (43.2)Medicare56 (27.7)35 (30.7)21 (23.9)Medicaid29 (14.36)12 (10.5)17 (19.3)Both Medicare and Medicaid18 (8.91)12 (10.5)6 (6.8)Other12 (5.94)6 (5.3)6 (6.8)Missing n 2 (0.99)2 (1.8)0 (0.0) Pain measures (Brief Pain Inventory) d Moderate to severe pain, n (%)128 (63.4)73 (64.0)55 (62.5)1.00Pain severity (range = 0–40)Mean (SD)4.3 (2.4)4.2 (2.4)4.4 (2.4)0.53Median (range)4.5 (0–10)4.8 (0–9.75)4.5 (0–10)0.53Missing n 1 (0.5)0 (0.0)1 (1.1)Pain interference (range = 0–70)Mean (SD)4.3 (2.8)4.3 (2.8)4.3 (2.8)0.97Median (range)4.3 (0–9.57)4.6 (0.9–57)4.1 (0,9–57)0.97Missing n 1 (0.5)0 (0.0)1 (1.1) Health measures General self‐rated health, n (%) e 0.07Excellent or very good20 (9.9)16 (14.0)4 (4.6)Good56 (27.7)32 (28.1)24 (27.3)Fair or poor122 (60.4)64 (56.2)58 (65.9)Missing n 4 (2.0)2 (1.8)2 (2.3) SF‐12 physical component summary score f Mean (SD)39.6 (10.9)39.9 (11.6)39.2 (10.0)0.62Below average (score < 50), n (%)160 (79.2)89 (78.1)71 (80.7)0.66Missing n 1 (0.5)0 (0.0)1 (1.1) SF‐12 mental component summary score f Mean (SD)45.1 (11.2)46.2 (10.8)43.7 (11.7)0.12Below average (score < 50), n (%)126 (62.4)66 (58.0)60 (68.2)0.14Missing n 1 (0.5)0 (0.0)1 (1.1)Abbreviations: EUC = enhanced usual care; SF‐12 = 12‐item Medical Outcomes Study Short Form; STOP = Subthreshold Opioid Use Disorder Prevention. a P‐values compare the EUC and STOP group using the chi‐square test (categorical variables) or t‐test (continuous variables) at each timepoint. P values < 0.05 are shown in bold. b Includes Asian, American Indian, Alaska Native, Native Hawaiian, Pacific Islander, multiracial and other races. c Categories included: student; keeping house; only temporarily laid off, sick leave or maternity leave; other (not specified). d Score ≥ 5 on the Brief Pain Inventory (BPI) short form, which measures pain severity (four items) and pain‐related functioning (seven items) using a 0–10 scale: 1–4 = mild pain; 5–6 = moderate pain; and 7–10 = severe pain [36]. e Single‐item measure of general self‐rated health (GSRH) from the SF‐12 [53]. f 12‐item Short‐Form Health Survey [37]. Demographic characteristics of participants by treatment condition, at baseline. Abbreviations: EUC = enhanced usual care; SF‐12 = 12‐item Medical Outcomes Study Short Form; STOP = Subthreshold Opioid Use Disorder Prevention. P‐values compare the EUC and STOP group using the chi‐square test (categorical variables) or t‐test (continuous variables) at each timepoint. P values < 0.05 are shown in bold. Includes Asian, American Indian, Alaska Native, Native Hawaiian, Pacific Islander, multiracial and other races. Categories included: student; keeping house; only temporarily laid off, sick leave or maternity leave; other (not specified). Score ≥ 5 on the Brief Pain Inventory (BPI) short form, which measures pain severity (four items) and pain‐related functioning (seven items) using a 0–10 scale: 1–4 = mild pain; 5–6 = moderate pain; and 7–10 = severe pain [36]. Single‐item measure of general self‐rated health (GSRH) from the SF‐12 [53]. 12‐item Short‐Form Health Survey [37]. At baseline, over half of participants in both groups (59.7% in EUC, 59.1% in STOP) reported risky opioid use in the 30 days prior to enrollment, which was primarily the non‐medical use of prescription opioids. In total, 116 had non‐medical prescription opioid use, 11 had illicit opioid use and seven had both. EUC participants reported 4.1 (SD = 6.6) mean days of risky opioid use; STOP participants reported 3.4 (SD = 5.7) days. Outcomes There were no significant differences between groups in the adjusted days of risky opioid use over the 180 days post‐enrollment (Table 2). The mean number of days of risky opioid use was 15.5 (SD = 32.6) in EUC and 12.2 (SD = 27.7) in STOP. This translates to a monthly average of 2.6 days versus 2.0 days of risky opioid use in EUC and STOP, respectively. The rate ratio comparing STOP versus EUC was 0.95 (95% CI = 0.52–1.74, P = 0.87). Most participants in both arms reported zero days of risky opioid use over a period of 180 days (Figure S2). TABLE 2Main results: summary of primary outcome analysis by treatment arm.Treatment armTotalEUCSTOP n = 202 n = 114 n = 88Days of risky opioid use a , mean (SD)14.1 (30.61)15.5 (32.64)12.2 (27.73) Treatment effect b Rate ratio c (95% CI) 0.95 (0.52–1.74) P 0.87Abbreviations: EUC = enhanced usual care; STOP = Subthreshold Opioid Use Disorder Prevention. a Days of risky opioid use measured in the 180 days after baseline. The mean and standard deviation are estimated using the patient participants who have no missing values of the response variable during months 1–6. b Results for the treatment effect are obtained from a fixed effects negative binomial model adjusted for site (stratification factor) and baseline value of the response. Missing values among the variables that contribute to the response value from months 1–6 and missing baseline values of the response are imputed via a multiple imputation approach. The missing data owing to intermittent missingness and owing to dropout are imputed separately assuming a missing at random (MAR) and a missing not at random (MNAR) mechanism, respectively. c The rate ratio is defined as the exponentiated estimate of the treatment effect as a ratio of the mean total number of days of risky opioid use for those enrolled to STOP versus EUC within the first 180 days post‐baseline. Main results: summary of primary outcome analysis by treatment arm. Abbreviations: EUC = enhanced usual care; STOP = Subthreshold Opioid Use Disorder Prevention. Days of risky opioid use measured in the 180 days after baseline. The mean and standard deviation are estimated using the patient participants who have no missing values of the response variable during months 1–6. Results for the treatment effect are obtained from a fixed effects negative binomial model adjusted for site (stratification factor) and baseline value of the response. Missing values among the variables that contribute to the response value from months 1–6 and missing baseline values of the response are imputed via a multiple imputation approach. The missing data owing to intermittent missingness and owing to dropout are imputed separately assuming a missing at random (MAR) and a missing not at random (MNAR) mechanism, respectively. The rate ratio is defined as the exponentiated estimate of the treatment effect as a ratio of the mean total number of days of risky opioid use for those enrolled to STOP versus EUC within the first 180 days post‐baseline. Secondary outcomes are presented in Table 3. A higher proportion of participants in EUC had moderate–severe OUD at 6 months (11.4% vs 1.1%, P = 0.02) and 12 months (5.3% vs 3.4%, P = 0.02). This difference was greatest at the 6‐month assessment, when 13 EUC participants met criteria for moderate–severe OUD, versus one STOP participant. At baseline, among the participants who reported receiving a prescription for opioid medication in the past 6 months, over one‐third (42.1% in EUC, 35.2% in STOP) had a COMM score indicating high‐risk prescription opioid use behavior [50]. The proportion with high‐risk prescription opioid use behavior declined in both groups at 6 and 12 months, and was significantly lower in the STOP condition at 12 months (16.7% in EUC vs 6.8% in STOP, P = 0.03). Non‐fatal overdose events were low at baseline and throughout the trial (with a median of zero overdose events in the past 6 months). Mean overdose risk behavior scores were significantly lower in STOP than in EUC at both 6 and 12 months. TABLE 3Substance use measured at baseline, 6 months and 12 months (n = 202).CharacteristicEUCSTOP P ** EUCSTOP P ** EUCSTOP P ** n = 114 n = 88 n = 114 n = 88 n = 114 n = 88 Risky opioid use, n (%) a Baseline (past 30 days) c 6 months (days 1–180) c 12 months (days 181–360) c Any opioid use b 68 (59.7)52 (59.1)0.8364 (56.1)38 (43.2)0.1347 (41.2)29 (33.0)0.32Mean days (SD^)4.1 (6.6)3.4 (5.7)0.4715.5 (32.6)12.2 (27.7)0.4610.1 (27.8)10.7 (30.1)0.89Median days (IQR*)2 (0–5)2 (0–4)3 (0–14)1 (0–10)0 (0–7.5)0 (0–5)Rx opioids65 (57.0)51 (58.0)0.726 (22.8)15 (17.1)0.4222 (19.3)13 (14.8)0.54Illicit opioids8 (7.0)3 (3.4)0.424 (3.5)2 (2.3)0.934 (3.5)3 (3.4)1.00Prescribed opioid in past 6 months86 (75.4)65 (73.9)0.9362 (54.4)43 (48.9)1.0048 (42.1)38 (43.2)0.73High risk for opioid misuse d 48 (42.1)31 (35.2)0.4121 (18.4)10 (11.4)0.3419 (16.7)6 (6.8) 0.03 Moderate–severe OUD, n (%) Baseline 6 months 12 months No or mild OUD (score 0–3)114 (100)88 (100)N/A*** 91 (79.9)73 (82.9) <0.001 96 (84.2)67 (76.2) <0.001 Moderate–severe OUD (score 4+)0 (0.0)0 (0.0)13 (11.4)1 (1.1)6 (5.3)3 (3.4) Days of other substance use a Baseline (Past 30 days) c 6 months (Days 1–180) c 12 months (Days 181–360) c Alcohol (heavy drinking) f Any use, n (%)36 (31.6)30 (34.1)0.8250 (43.9)32 (36.4)0.4847 (41.2)30 (34.1)0.38Mean days (SD)3.3 (7.5)2.5 (6.4)0.439.9 (23.0)7.3 (18.1)0.4111.2 (28.5)7.0 (21.8)0.26Median (IQR)0 (0–2)0 (0–2)0 (0–8)0 (0–3.5)0 (0–7.5)0 (0–1.5) Cannabis Any use, n (%)39 (34.2)33 (37.5)0.7448 (42.1)32 (36.4)0.7151 (44.7)39 (44.3)1.00Mean days (SD)5.6 (10.5)5.3 (9.6)0.8431.0 (55.2)34.3 (61.5)0.7134.0 (58.4)33.2 (61.2)0.93Median days (IQR)0 (0–4.8)0 (0–5.0)0 (0–26.5)0 (0–37.0)0 (0–33.5)0 (0–21.0) Benzodiazepines Any use, n (%)16 (14.0)12 (13.6)1.0019 (16.7)9 (10.2)0.3414 (12.3)12 (13.6)0.93Mean days (SD)0.7 (3.1)0.8 (2.6)0.964.0 (18.7)2.4 (11.1)0.473.0 (16.4)2.9 (10.8)0.93Median days (IQR)0 (0,0)0 (0,0)0 (0,0)0 (0,0)0 (0,0)0 (0,0) Stimulants Any use, n (%)5 (4.4)3 (3.4)1.008 (7.0)3 (3.4)0.498 (7.0)2 (2.3)0.23Mean (SD)0.1 (0.7)0.8 (4.6)0.180.6 (3.5)1.2 (8.1)0.540.9 (4.3)2.0 (15.4)0.53Median (IQR)0 (0–0)0 (0–0)0 (0–0)0 (0–0)0 (0–0)0 (0–0) Overdose risk behavior g Overdose risk score (range 0–34)0.33 0.03 0.02 Mean (SD)3.3 (3.3)2.8 (3.0)2.1 (3.1)1.2 (2.0)2.1 (3.4)1.2 (2.1)Median (IQR)3 (1–5)2 (0.5–4)0 (0–3)0 (0–2)0 (0–3)0 (0–1.75)Nonfatal overdose events in past 6 mos0.520.990.99Mean (SD)0.9 (2.5)1.6 (10.0)0.7 (4.5)0.7 (5.9)0.3 (1.1)0.3 (1.7)Median (IQR)0 (0–0)0 (0–0)0 (0–0)0 (0–0)0 (0–0)0 (0–0) Readiness to change substance use h Readiness score (range 1–10)0.460.44NAMean (SD)6.1 (3.2)5.7 (4.0)5.7 (3.2)5.3 (3.6)Median (IQR)6 (3,9)5 (1–10)5 (3–9)5 (2–10)Abbreviations: EUC = enhanced usual care; OUD = opioid use disorder; Rx = prescription; STOP = Subthreshold Opioid Use Disorder Prevention.** P‐values compare the EUC and STOP groups using the chi‐square test (categorical variables) or t‐test (continuous variables) at each timepoint. P values < 0.05 are shown in bold.*** P‐value cannot be calculated because no individuals had moderate–severe OUD at baseline, consistent with the eligibility criteria.Note that percentages may not sum to 100% owing to missing data. a Days of substance use: calculated from monthly surveys asking about use in the past 30 days using single‐item questions based on the Addiction Severity Index and Current Opioid Misuse Measure. The survey is described in Liebschutz JM, Subramaniam GA, Stone R, Appleton N, Gelberg L, Lovejoy TI, Bunting AM, Cleland CM, Lasser KE, Beers D, Abrams C, McCormack J, Potter GE, Case A, Revoredo L, Jelstrom EM, Kline MM, Wu LT, McNeely J. Subthreshold opioid use disorder prevention (STOP) trial: a cluster randomized clinical trial: study design and methods. Addict Sci Clin Pract. 2023;18(1):70. doi:10.1186/s13722‐023‐00424‐8. PMID: 37980494; PMCID: PMC10657560. b Any opioid use includes illicit opioid use and non‐medical prescription opioid use. c Risky opioid use defined as illicit opioid use and/or non‐medical prescription opioid use in the past 3 months measured on the TAPS Tool, and/or non‐medical use of prescription opioids (taking more opioid medication than prescribed or opioid medications belonging to someone else) as reported on the Current Opioid Misuse Measure (COMM). d Score ≥9 on the Current Opioid Misuse Measure (COMM), a 17‐item self‐administered questionnaire, where a score of 9 or higher indicates high risk. The COMM was only administered to participants who self‐reported the use of prescribed opioid medications. Butler SF, Budman SH, Fanciullo GJ, Jamison RN. Cross validation of the current opioid misuse measure to monitor chronic pain patients on opioid therapy. Clin J Pain. 2010;26(9):770–6. e Measured with the Composite International Diagnostic Interview (CIDI) modified for DSM‐5, applying the standard cut‐off of 4+ criteria for moderate–severe OUD. f Heavy drinking days defined as days consuming 4+ drinks for women or 5+ drinks for men. g Non‐fatal overdose questionnaire: a nine‐item self‐administered questionnaire measuring the frequency of overdose risk behaviors in the past 6 months. The total risk score is an aggregation of responses to individual questions; range = 0–34. Bohnert ASB, Walton MA, Cunningham RM, Ilgen MA, Barry K, Chermack ST, et al. Overdose and adverse drug event experiences among adult patients in the emergency department. Addict Behav. 2018;86:66–72. h Collected at baseline and 6 month assessments, using the item ‘On a scale of 1 to 10, how ready are you to change your alcohol, opioid, or other drug use?’; range = 1–10, with 10 being highest readiness. Williams, E.C., et al., Do brief measures of readiness to change predict alcohol consumption and consequences in primary care patients with unhealthy alcohol use? Alcohol Clin Exp Res. 2007;31(3):428–35. Substance use measured at baseline, 6 months and 12 months (n = 202). Abbreviations: EUC = enhanced usual care; OUD = opioid use disorder; Rx = prescription; STOP = Subthreshold Opioid Use Disorder Prevention. P‐values compare the EUC and STOP groups using the chi‐square test (categorical variables) or t‐test (continuous variables) at each timepoint. P values < 0.05 are shown in bold. P‐value cannot be calculated because no individuals had moderate–severe OUD at baseline, consistent with the eligibility criteria. Note that percentages may not sum to 100% owing to missing data. Days of substance use: calculated from monthly surveys asking about use in the past 30 days using single‐item questions based on the Addiction Severity Index and Current Opioid Misuse Measure. The survey is described in Liebschutz JM, Subramaniam GA, Stone R, Appleton N, Gelberg L, Lovejoy TI, Bunting AM, Cleland CM, Lasser KE, Beers D, Abrams C, McCormack J, Potter GE, Case A, Revoredo L, Jelstrom EM, Kline MM, Wu LT, McNeely J. Subthreshold opioid use disorder prevention (STOP) trial: a cluster randomized clinical trial: study design and methods. Addict Sci Clin Pract. 2023;18(1):70. doi:10.1186/s13722‐023‐00424‐8. PMID: 37980494; PMCID: PMC10657560. Any opioid use includes illicit opioid use and non‐medical prescription opioid use. Risky opioid use defined as illicit opioid use and/or non‐medical prescription opioid use in the past 3 months measured on the TAPS Tool, and/or non‐medical use of prescription opioids (taking more opioid medication than prescribed or opioid medications belonging to someone else) as reported on the Current Opioid Misuse Measure (COMM). Score ≥9 on the Current Opioid Misuse Measure (COMM), a 17‐item self‐administered questionnaire, where a score of 9 or higher indicates high risk. The COMM was only administered to participants who self‐reported the use of prescribed opioid medications. Butler SF, Budman SH, Fanciullo GJ, Jamison RN. Cross validation of the current opioid misuse measure to monitor chronic pain patients on opioid therapy. Clin J Pain. 2010;26(9):770–6. Measured with the Composite International Diagnostic Interview (CIDI) modified for DSM‐5, applying the standard cut‐off of 4+ criteria for moderate–severe OUD. Heavy drinking days defined as days consuming 4+ drinks for women or 5+ drinks for men. Non‐fatal overdose questionnaire: a nine‐item self‐administered questionnaire measuring the frequency of overdose risk behaviors in the past 6 months. The total risk score is an aggregation of responses to individual questions; range = 0–34. Bohnert ASB, Walton MA, Cunningham RM, Ilgen MA, Barry K, Chermack ST, et al. Overdose and adverse drug event experiences among adult patients in the emergency department. Addict Behav. 2018;86:66–72. Collected at baseline and 6 month assessments, using the item ‘On a scale of 1 to 10, how ready are you to change your alcohol, opioid, or other drug use?’; range = 1–10, with 10 being highest readiness. Williams, E.C., et al., Do brief measures of readiness to change predict alcohol consumption and consequences in primary care patients with unhealthy alcohol use? Alcohol Clin Exp Res. 2007;31(3):428–35. The most frequently used non‐opioid substances were alcohol (heavy drinking) and cannabis. At 6 months, the mean number of heavy drinking days was 9.9 (SD = 23.0) in EUC and 7.3 (SD = 18.1) in STOP (P = 0.41). The mean number of cannabis use days at 6 months was 31.0 (SD = 55.2) in EUC and 34.3 (SD = 61.5) in STOP (P = 0.71). The mean scores for readiness to change substance use (1–10 scale) were 5.7 (SD = 3.2) and 5.3 (SD = 3.6) in EUC and STOP, respectively, at 6 months (P = 0.44). Supplemental analyses of the primary outcome Planned supplemental analyses of the primary outcome are included in Supporting information. These include: sensitivity analyses of missing data mechanisms assuming MAR for both intermittent and dropout missingness, and a complete case analysis (Appendix S1); individual‐level covariate adjusted analysis (Appendix S2); subgroup analyses for sex, age, race and ethnicity (Appendix S3); per‐protocol analysis excluding 53 patients that may have had visits with a PCP in the opposite treatment condition or received the incorrect video (Appendix S4); and zero‐inflated negative binomial model (Appendix S5). For all supplemental analyses, results were similar to the main primary outcome analysis, with no statistically significant differences between groups. Table S2 shows participant engagement in the intervention, for those in the STOP arm. A majority (60.6%) of participants received at least one session of each intervention component. Brief advice was delivered by the PCP to 62% of participants, and 93% received the brief advice video. Most (94%) had at least one contact with the NCM, and the mean number of contacts during the intervention period was 23.4 (SD = 22.2). Almost all (98%) received at least one health coaching session; 91% participated in at least the two required sessions, and 57% completed the maximum of six sessions. On the assessment of blinding completed at the end of the study, one‐third of participants in each arm reported believing the study was focused on substance use only [n = 29 (33%) in STOP, n = 38 (33%) in EUC], and approximately one‐quarter believed the study was focused on both substance use and healthy lifestyle [n = 25 (28%) in STOP, n = 27 (24%) in EUC]. Safety events In this minimal risk study, causally related adverse and serious adverse events were not anticipated. No safety events were linked to the intervention. The two deaths that occurred during the trial were unrelated to the intervention, and neither death was the result of an overdose. DISCUSSION This multi‐site cluster randomized clinical trial offers a rare look at the substance use patterns and trajectories of adult primary care patients with risky opioid use who did not have moderate–severe OUD at study entry. An adapted collaborative care model did not significantly reduce the number of days of risky opioid use at 6 months, compared with EUC. Days of risky opioid use were lower than expected throughout the study period. Mean days of risky opioid use were low at baseline (average of 3.8 days in the past month) and declined in both the intervention and control groups, with minimal differences between the groups. Despite the overall low level of risky opioid use, our findings indicate that this patient population may be at risk for developing moderate–severe OUD. At the 6‐month assessment, 13 EUC participants and one STOP participant (6.9% of the sample) had developed moderate–severe OUD. Overdose risk behavior scores were significantly lower for those in the STOP condition compared with EUC, with low non‐fatal overdose rates and risk behaviors overall. While these findings should be interpreted with caution given the limitations of our sample, they do offer some insight into a neglected area of research on secondary prevention interventions for patients with risky opioid use. Longitudinal studies with larger samples are needed to provide a more comprehensive understanding of this population’s patterns of use and ongoing OUD and overdose risk. Participants had high rates of non‐opioid substance use as well as physical and mental health comorbidities, which may pose challenges to reducing opioid use [54]. Prior studies have found that chronic pain is frequently intertwined with substance use disorders [55, 56] and mental health conditions [57]. However, opioid dose reduction trials have not reported high rates of other substance use [58]. Given the complex health conditions of our enrolled patient population and the relatively low days of risky opioid use, we posit that providers and participants prioritized other concerns over opioid reduction. A greater focus on preventing undesirable outcomes related to risky opioid use, such as physical and mental health consequences or the development of moderate–severe OUD, rather than reducing the number of days of use alone, could offer a productive approach for future interventions. Our trial occurred during a period of decreased opioid prescribing [59, 60, 61]. At all sites, study teams encountered challenges with identifying patients who met the criteria of having risky opioid use without meeting the criteria for moderate–severe OUD. Of the >42 000 anonymous pre‐screens completed, 98% were ineligible owing to reporting no current risky opioid use. Based on prior studies, we had anticipated higher rates of risky opioid use among adult primary care patients. In screening studies conducted during 2011–2016, 3.9%–6.6% of adult primary care patients had illicit opioid use [10, 12]; among patients on chronic opioid therapy, 30% had non‐medical opioid use [62]. But recent studies, including this trial, have found low rates of opioid use on screening questionnaires delivered in primary care [63, 64]. With increased vigilance around opioid prescribing, medical patients may be reluctant to report drug use during screening [65], though we attempted to minimize this bias by using an anonymous self‐administered approach. Future studies and clinical programs seeking to engage this population may need to expand the reach of interventions beyond primary care (e.g. to pharmacy or community‐based settings) and build trust with patients who are fearful of the consequences of reporting risky opioid use to medical providers [66, 67, 68]. The study team specifically masked the purpose of the trial because of this fear and would recommend this for future research. Clinical programs seeking to provide early intervention for patients with risky opioid use should directly address patient concerns about the potential consequences of reporting substance use, and consider taking a holistic approach to discussing opioid use in the context of other health conditions that may be less stigmatized (e.g. managing pain or mood symptoms). Integrating substance use services with programs addressing other medical and mental health conditions (e.g. disease management and behavioral health care management programs) could help to reduce stigma and increase the feasibility of implementing and sustaining interventions for the relatively small population of patients with risky opioid use in primary care. Despite the recruitment challenges, 83% of patients who screened eligible agreed to enroll and had high levels of assessment completion. Those in the STOP condition were highly engaged with NCMs and health coaches, indicating that collaborative care may be acceptable to primary care patients with risky opioid use. The trial was conducted during a time when primary care was disrupted by the COVID‐19 pandemic [69, 70], and one‐third of intervention patients did not receive brief advice from the PCP. This may be through the lack of a scheduled appointment within that time, competing demands during the clinic session, poor documentation or PCP discomfort. The strengths of the trial included diverse practice types and study locations, high rates of participation in study assessments from enrolled patients and minimal missing data for the primary outcome. Study limitations included lower than anticipated days of opioid use in both groups and failure to reach the original target sample size of 480 participants [9]. The power calculations assumed a difference of 10–12 days in the number of days of risky opioid use between the groups for the primary outcome measure, and indicated that a sample of 200 participants would have been adequate to detect this difference. A larger sample would have been required to detect smaller effect sizes, but a difference of fewer than 10 days over 6 months of measurement (i.e. an average of 1–2 days per month) was not deemed clinically meaningful. More individuals in our study sample reported zero days of opioid use, with greater variation in the number of days of risky opioid use, than was anticipated in our sample size calculations. The large variation in the number of days of risky opioid use suggests this may be a highly variable outcome that either needs refinement or more complex modeling to reduce unexplained variation in future studies. The primary outcome analysis adjusted for baseline days of risky opioid use, but not for OUD severity (participants could enter the study if they had CIDI scores of 0–3, indicating no or mild OUD). There were some missing data for the moderate–severe OUD measure, which was not equally distributed between the EUC and STOP groups, and the low number of cases developing moderate–severe OUD limited our analyses of this secondary outcome. Twelve months may also be too brief a timescale to measure the prevention of moderate–severe OUD and other opioid‐related health outcomes. There were missing data on PCP‐delivered brief advice for the first 22 consecutively enrolled participants because of an error in the study forms. Questions assessing the primary outcome were based on previously validated measures [48, 49, 50, 51], but have not themselves been validated. We also cannot rule out assessment reactivity to monthly surveys and social desirability bias, which could have diminished differences between EUC and intervention conditions. Assessment reactivity has impacted prior studies of risky substance use in primary care populations [10, 23, 44, 71]. We used several procedures to mitigate these effects, including the use of self‐administered assessments [72, 73, 74], blinding participants to the study focus on opioid use, presenting questions about opioid and other substance use alongside other health behaviors, and accounting for assessment reactivity in our power analysis. The patterns of change in number of days of use were not uniform across substances, which suggests that social desirability and assessment effects did not substantially impact our findings, but the study did not include a minimal assessment sample. CONCLUSION Compared with EUC, the STOP collaborative care intervention—employing PCP brief advice, NCM and remote health coaching support—did not result in greater reductions in the number of days of risky opioid use among this sample of primary care patients. However, fewer participants in the STOP group, compared with EUC, developed moderate–severe OUD during the trial. The patients’ high comorbidity burden may have hindered efforts to reduce risky opioid use, and unexpectedly low levels of risky opioid use made it challenging to detect differences between groups. The high rates of engagement with health coaches and NCMs indicate patient receptivity to receiving services that address opioid use along with other complex health needs in primary care settings. Future longitudinal studies could target specific and longer interventions for risky opioid use and assess their impact on preventing worsening OUD and related health outcomes that may take longer to manifest. AUTHOR CONTRIBUTIONS Jennifer McNeely: Conceptualization; investigation; methodology; resources; writing—review and editing; writing—original draft; visualization; supervision; project administration; funding acquisition. Geetha A. Subramaniam: Conceptualization; methodology; investigation; resources; writing—review and editing; visualization. Rebecca Stone: Methodology; investigation; resources; writing—review and editing; visualization; supervision; project administration. Shayna Mazel: Writing—review and editing; validation; project administration; data curation; formal analysis. Noa Appleton: Investigation; methodology; writing—review and editing; project administration; supervision; resources; visualization. Yasna Rostam‐Abadi: Methodology; writing—review and editing; project administration. Amanda M. Bunting: Methodology; investigation; writing—review and editing; project administration. Yidan Shi: Data curation; software; formal analysis; validation; writing—review and editing. Travis I. Lovejoy: Project administration; conceptualization; methodology; investigation; writing—review and editing. Lillian Gelberg: Investigation; conceptualization; methodology; writing—review and editing; project administration. Donna Beers: Methodology; writing—review and editing; supervision. Margaret Kline: Methodology; investigation; software; formal analysis; validation; writing—review and editing; resources; supervision. Song Zhang: Investigation; methodology; validation; writing—review and editing; formal analysis; software; supervision; resources. Tobie Kim: Investigation; methodology; validation; writing—review and editing; software; formal analysis; supervision; resources. Ashley Case: Investigation; methodology; validation; writing—review and editing; formal analysis; software; supervision; resources. Jennifer McCormack: Investigation; methodology; validation; writing—review and editing; software; formal analysis; supervision; resources. Michael A. Incze: Investigation; writing—review and editing; supervision. Jane M. Liebschutz: Conceptualization; methodology; investigation; writing—original draft; writing—review and editing; visualization; supervision; project administration; funding acquisition; resources. DECLARATION OF INTERESTS This article reflects the views of the authors and may not reflect the opinions, views and official policy or position of the National Institutes of Health, US Department of Health and Human Services, Department of Veterans Affairs, or any of its affiliated institutions or agencies. G.A.S. was substantially involved in this project, consistent with her role as Scientific Officer on cooperative agreements (UG1DA013035, UG1DA049436, UG1DA015815, UG1DA040317, UG1DA049435). G.A.S. has no conflicts to report. J.M. is a consultant for Eli Lilly and Company and is a member of the Coordinating Committee for the International Network on Brief Interventions for Alcohol and Other Drugs (INEBRIA), a voluntary not‐for‐profit research network. J.M.L. is a consultant for Biomotivate and Kenvue. A.J.G. receives an annual honorarium for an online chapter on alcohol management in the perioperative period from the UpToDate online reference. In the last 3 years, A.J.G. has been on the board of directors for the American Society of Addiction Medicine (ASAM), the Association for Multidisciplinary Education and Research in Substance use and Addiction (AMERSA) and the International Society of Addiction Journal Editors (ISAJE), all not‐for‐profit organizations. A.J.G. does not receive remuneration for these duties. A.J.G. receives remuneration from AMERSA for being the Editor‐in‐Chief of their peer‐reviewed journal. A.J.G. does not believe that these disclosures reflect any conflict to this article. CLINICAL TRIALS REGISTRATION Registry Name: Subthreshold Opioid Use Disorder Prevention (STOP) Trial. Submission Date (first submitted): 23 December 2019. Trial ID: NCT04218201. URL: https://clinicaltrials.gov/study/NCT04218201. Supporting information Appendix S1. Sensitivity analyses for missing data assumptions. Appendix S2. Individual‐level covariate adjusted analysis. Appendix S3. Subgroup analyses for sex, age, race and ethnicity. Appendix S4. Per‐protocol population analysis. Appendix S5. Primary outcome analysis using zero inflated negative binomial models. Figure S1. Primary care provider (PCP) enrollment. Figure S2. Number of opioid use days during the primary outcome measurement period, by treatment arm.1 Table S1. Study sites and participating clinics. Table S2. Receipt of intervention by participants in the STOP condition (n = 88). STOP TRIAL SITE PRINCIPAL INVESTIGATORS Karen Alexander 14, Seuli Bose Brill 15, Sarah Kawasaki 16, Adam J Gordon 17, 18, Hyunouk Hong 19. ACKNOWLEDGEMENTS The authors would like to thank the CTN‐0101 Research Coordinators, Research Assistants, Practice Managers, Nurse Care Managers and Telephone Health Coaches for their diligent and impactful work with study participants. We also recognize the efforts of Clinical Champions Martin Fried, Jared Moore and Lauren Vaszil. Thank you also to Fei Guo, Michael Otterstatter, Charles Cleland, Leslie Revoredo and Gail Potter for their important scientific contributions. McNeely J, Subramaniam GA, Stone R, Mazel S, Appleton N, Rostam‐Abadi Y, et al. Collaborative care intervention for risky opioid use among primary care patients: The STOP randomized clinical trial. Addiction. 2026;121(10):2807–2821. 10.1111/add.70504 Funding information This study was supported by the National Institute on Drug Abuse Clinical Trials Network (NIDA‐CTN): Appalachian Node (UG1DA049436); New York Node (UG1DA013035); Western States Node (UG1DA015815); Southern California Node (UG1DA049435); Ohio Valley Node (UG1DA013732); Northeast Node (UG1DA040309); Greater Intermountain Node; (UG1DA049444). Infrastructure support for author A.J.G. was provided, in part, by the Greater Intermountain Node (GIN; NIH/NIDA 1UG1DA049444) of the National Institute on Drug Abuse Clinical Trials Network and the Department of Veterans Affairs Health Services Research and Development Service Informatics, Decision‐Enhancement, and Analytic Sciences (IDEAS; CIN 13‐414) Center of Innovation. Contributor Information Jennifer McNeely, Email: jennifer.mcneely@nyulangone.org. the STOP trial site principal investigators: Karen Alexander, Seuli Bose Brill, Sarah Kawasaki, Adam J. Gordon, and Hyunouk Hong DATA AVAILABILITY STATEMENT All individual participant data collected during the trial, after de‐identification and with permission from the lead investigator, will be accessible to anyone who wishes to access the data. Data are available indefinitely (https://datashare.nida.nih.gov). Data sets for CTN protocols will be available after the primary article has been accepted for publication or when the data are locked for more than 18 months, whichever comes first. REFERENCES Associated Data Supplementary Materials Appendix S1. Sensitivity analyses for missing data assumptions. Appendix S2. Individual‐level covariate adjusted analysis. Appendix S3. Subgroup analyses for sex, age, race and ethnicity. Appendix S4. Per‐protocol population analysis. Appendix S5. Primary outcome analysis using zero inflated negative binomial models. Figure S1. Primary care provider (PCP) enrollment. Figure S2. Number of opioid use days during the primary outcome measurement period, by treatment arm.1 Table S1. Study sites and participating clinics. Table S2. Receipt of intervention by participants in the STOP condition (n = 88). Data Availability Statement All individual participant data collected during the trial, after de‐identification and with permission from the lead investigator, will be accessible to anyone who wishes to access the data. Data are available indefinitely (https://datashare.nida.nih.gov). Data sets for CTN protocols will be available after the primary article has been accepted for publication or when the data are locked for more than 18 months, whichever comes first.