Messages include similar components varied throughout the 30-day messaging period

Health coaches elicit how participants manage challenges to build self-efficacy based on prior success and explore barriers, and use rulers to elicit confidence and commitment to change by elaborating change talk. The health coaches close with a strategic summary, eliciting a next step toward goals. Health coaches review tailored resources, such as obtaining naloxone, and encourage disposal of leftover medications with the provided pill disposal bags. We developed a web-based portal with manualized content for future implementation into existing means of patient communication . The portal resides on internal research servers in the Michigan Medicine network and is only accessible to study staff and enrolled participants who have created personalized passwords. Health coaches push a portal message about twice weekly using MI strategies to engage participants in a dialogue around developmentally tailored and personally relevant topics consistent with the Why and How model of MI. Messages are tailored using baseline data . For those receiving the health coach session, coaches review a progress note from the session for further tailoring messages . Messages are further tailored to AYA responses, as done in-the-moment during face-to-face therapy. For example, messages about tools are tailored for an individual who reports clinically significant depression symptoms at baseline ,procona valencia whereas those who report no depression may receive tools related to general stress reduction coping skills .

Also, messages about substance use are tailored, for example, focusing on benefits of avoiding concurrent use of alcohol/opioids for those who drink and overdose response for those with more severe misuse. Although participants may not respond to every message sent, simply viewing a message could contribute to behavior change. Further, token gifts displaying the portal name are sent as a participation reminder during the 30-day period, but are not a contingency of participation. When participants do not reply to an initial message, health coaches send follow-up messages in the portal with reminders sent via other modalities or a telephone call. In addition, health coaches respond to participants’ replies in an MI manner to elaborate the conversation and elicit and reinforce change talk, minimize sustain talk, and plan. The participant-facing portal provides a crisis text/phone line for immediate help and link to online resources.Given the lack of opioid focused screening and prevention interventions or early interventions for AYAs in the ED setting, we chose an EUC as a control condition to offer a minimal resources. The EUC condition involves reviewing a community resource brochure, exceeding the ED’s current standard of care, at the intake . For post-intake EUC, we send the community resource brochure to participants by email. The resources include information on topics such as: storage/disposal, overdose prevention, naloxone, suicide hotlines, mental health, and substance use treatment. Although pregnant women are excluded at baseline, enrolled participants may become pregnant during the study, so risks of opioid/other substance use during pregnancy are included in this resource brochure.

The EUC resource brochure is additionally shared with all participants at enrollment and in the form of a pdf included with each follow-up survey invitation, regardless of condition, to control for any effects of exposure across conditions. Further, as part of EUC, all participants receive an opioid disposal bag when enrolling, either in-person or mailed. The trial is part of a cooperative along with several other studies funded through the NIH HEAL Initiative. As part of the HPC, investigators from each site agreed to use several common measures across trials. As such, several measures were adapted to meet unique needs of the HPC studies and will be cited herein as “HPC” when significant modifications were made for use by the HPC. Measures of outcome are repeated across screening/baseline and follow-up assessments. Below, we focus on our pre-registered primary and secondary outcomes, however, our assessments also include a number of other measures of potential mediators and moderators, and other exploratory outcomes. The screening survey is self-administered and contains a number of items to assess trial eligibility and stratification variables as well as selected demographics, consistent with the HPC. Trial eligibility involve: a) past 12-month prescription or illicit opioid misuse, or b) past 12-month prescription opioid use plus at least one other risk factor. The specific risk factors are: other drug use or misuse of prescription sedatives or stimulants in the past 3 months; binge drinking in the past 3 months; positive 2-week depression screening; past-year suicide attempt; or past 2-week suicidal ideation.

Items assessing opioid use and misuse are based on definitions from the HPC, with response options based on the National Epidemiological Survey of Alcohol and Related Conditions capturing frequency on a 12-point scale from “Never” to “More than Once a Day.”Other drugs queried are: cannabis, cocaine, methamphetamine, and hallucinogens, with a similar 3-month frequency response scale . Misuse of prescription sedatives and stimulants use the same 3-month response scale, with misuse defined consistent with the HPC-provided wording above. Binge drinking is assessed within the Alcohol Use Disorders Identification Test-Consumption , modified for a 3-month period and sex-specific binge levels . We use the Patient Health Questionnaire-2 to screen for past 2-week depression symptoms.Recent suicide ideation is captured using an item within a self-reported Columbia-Suicide Severity Rating Scale severity of ideation sub-scale, with a single item assessing past-year suicide attempts adapted from the C-SSRS behavior scale. In addition, to screen for exclusion criteria and to determine opioid risk stratification, we measure 3-month misuse of both prescription and illicit opioids per the HPC definitions combined with the NIDA-Modified ASSIST .Individuals with scores of 27+ on either ASSIST sub-scale are excluded and we also exclude individuals who report injection drug use on the ASSIST item. Prior to approach ED patients who are presenting for pregnancy or related reasons or cancer are excluded from recruitment; however, we also query pregnancy status and cancer status, to assess these exclusion criteria. We will compare the effects of the health coach session+EUC, EUC+portal, and combination of the health coach session+portal interventions to EUC+EUC on the primary outcome of opioid misuse severity over time for four measurements in longitudinal analyses. Additionally, we will examine the comparative effectiveness of the interventions. Generalized linear mixed models , also known as random effects or growth curve models, will be used to analyze the longitudinal data with a log link. GLMMs use all available measurements, allowing subjects to have an unequal number of observations and producing unbiased parameter estimates as long as unobserved values are missing completely at random or missing at random . The model will include fixed effects for the effects of health coach session and portal and the interaction between health coach session and portal. Additionally, the model will include a fixed effect for time point and interactions between time points and each main effect and interaction term of the interventions. With this model,flower bucket we will be able to assess the main effects of the health coach session and portal and the interaction effect of health coach session x portal at each time point. We are primarily interested in the pairwise comparisons of treatment combinations. The GLMM will also include random effects for the intercept and time and an unstructured within-person correlation structure for the residual errors and will adjust for age and sex. Model diagnostics will be used to determine suitability of more parsimonious correlation structures, and nonlinear effects for time. Additionally, we will assess the fit under the Poisson distribution assumption using the scaled Pearson statistic and compare to the fit of the over-dispersed Poisson, negative binomial, zero-inflated Poisson, and/or zero-inflated binomial models by log likelihood values. Secondary analyses examining efficacy on other outcomes will be modeled similar to above. The Poisson distribution will be checked for these outcomes where appropriate whereas the identity link will be used for continuous outcomes and the logit link for binary outcomes. In addition, we will examine baseline and time-varying factors that predict outcomes. To investigate moderation, interactions between the moderators of interest and main effects of the treatment variable will be assessed in the models specified above.

To investigate mediation, we will establish the three preconditions for mediation derived from Baron and Kenny’s causal steps approach. We will examine the mediators in structural equation models using the R package lava an to determine indirect effects using bootstrapping. Bootstrapping does not assume normality of the product term used to examine indirect effects. Our mediation hypotheses are not time specific, so we will compute and report all indirect pathways and their respective effect size coefficients. The reported path coefficients are completely standardized. The reported overall total effect, overall direct effect and overall indirect effect coefficients are unstandardized. We will use the proposed cut-off criteria to assess the fit between hypothesized models and the data: CFI>0.95, RMSEA<0.06, SRMR<0.08. The lava an package is able to use full information maximum likelihood estimation to efficiently address any missing data that is either missing completely at random or missing at random in any of these constructs. Power and sample size was estimated based on prior work from our team, the brief intervention literature, and initial pilot data which showed lower opioid misuse base rates than prior work, although effect sizes from pilot studies can have large standard errors and be unstable.Sample size for this study is based on the primary aim, with opioid misuse score as the primary outcome. We are powered for our primary aim to compare each of the three intervention groups to EUC+EUC and to each other. Power was estimated based on N = 1,170 and an 85% follow up rate which does not consider imputations and other strategies for handling missing data without reducing sample size. We estimated power assuming a simpler model with one follow-up, a Poisson distribution of the primary outcome, and computed sample size by simulation using R 3.5.1. Conservatively, we used a Bonferroni correction for 6 pairwise treatment comparisons, setting the type I error at 0.008. Then, with 248 participants/group, we maintain 90% power to detect a rate ratio of 0.819 if the base rate is as low as 2.8; if the base rate is as low as 1.5, we have 90% power to detect a rate ratio of 0.73. These effect sizes are modest, however, the interventions are scalable and thus could have high impact. Scalable, developmentally-tailored efficacious strategies are needed to not only prevent misuse among those using opioids, especially those with other risks associated with adverse outcomes, but also to prevent transition to opioid use disorder. If our interventions are deemed efficacious, this RCT will provide such scalable early interventions for health systems using technology infrastructure that already exists. For example, in response to the COVID-19 pandemic telehealth is now more common to health systems than ever before68–71, meaning there is now infrastructure in many health systems to deliver such telemedicine driven to ED patients. Further, patient portals are being used throughout health systems in a variety of ways72, but have not yet been tapped for their potential to deliver behavioral health interventions. The 24/7 access to patient portals is an advantage because participants can interact when they choose to in their busy lives from a variety of locations. Moreover, these interventions also provide a flexible working model for staff, who could be located in the ED, in a separate telemedicine hub, or at home. For example, in the context of our pilot work conducted early during the COVID-19 pandemic, interventions occurred remotely from health coaches to patients’ in their homes after the ED visit, an even more flexible model. Our interventions are also consistent with continuing care approaches73 which facilitate linkage to other healthcare settings. Harnessing technology for resource-light delivery could therefore maximize translation of these secondary prevention efforts into routine clinical care, with high impact on AYAs’ trajectories of health and well-being.This study has several innovative elements. Although single-session brief motivational interventions are not uniquely novel, the application to opioid misuse and our delivery approach increase innovation by harnessing technology, which is appealing to AYAs, to facilitate remote video delivery by a health coach in a tele-medicine hub, promoting fidelity using an online clinician support toolkit to structure the session while allowing personalization. Online support tool kits such as ours allow for within session data capture potentially identifying active ingredients of interventions.Further, extending intervention delivery post-discharge and capitalizing on young peoples’ use of technology, our portal messaging is novel, and, with the rise of patient portals in health systems, this feature enhances future ED implementation and extension of interventions into AYAs’ day to-day lives.