We conducted a systematic search of electronic databases including CINAHL, Medline, PsychInfo, and Scopus with the assistance of a research librarian to identify English language abstracts of studies published January 2013-May 2018 for inclusion in the review. Search terms included those associated with opioids (e.g., analgesics, opioid, opiate, painkiller, pain reliever, heroin), key outcomes of interest (e.g., overdose, death, abuse, misuse), and intervention strategies (e.g., prescribing, guideline, legislation, education, naloxone) (see Supplementary Appendix: Search Strategy).

The authors searched the reference lists of identified publications to find studies not detected in the search; when potentially relevant publications were identified, the authors screened abstracts to determine appropriateness for inclusion. Authors nominated publications for inclusion that were otherwise known based on experience (N = 23); such studies could have been published after the end date of the electronic abstract search up until the review activities were completed near the end of 2018 to allow for the most up-to-date review as practical.

Selection criteria included consideration of publications including studies that were part of the previous review (Haegerich et al., 2014), evaluated interventions intended to prevent opioid overdose, conducted in the United States, printed in English language, and with rigorous designs such as experimental or quasi-experimental designs (e.g., RCT, time series, pre-post, or pre-post with comparison). Non-comparative studies were included or retained from the original systematic review (Haegerich et al., 2014) if they represented a novel intervention with promise but with a small number of studies with rigorous designs evaluating them to inform effectiveness (e.g., naloxone distribution). Some studies included in the previous review are not included within this review if they were non-US studies or focused solely on drugs other than opioids (e.g., benzodiazepines).

Included interventions were determined to have systems-level implementation feasibility through state and local health agencies or collaborations with health systems. Evaluations must have included reports of desired outcomes, including provider behavior (e.g., opioid prescribing, guideline-concordant care), patient behavior (e.g., use of multiple providers or pharmacies, opioid misuse), and health outcomes (e.g., overdose from prescription or illicit opioids). The term “patient” was used to reflect findings relevant to the general public. When publications of educational interventions assessed knowledge only, the study was included, and the outcome was coded as “behavior”.

Consistent with our prior systematic review, we excluded biochemical and animal studies; case reports; and evaluations of abuse-deterrent formulations, opioid use disorder treatment (e.g., methadone, buprenorphine, or naltrexone), and compulsory drug treatment/drug courts. Medications for opioid use disorder have been demonstrated to save lives (NASEM, 2019). Evaluations of public health approaches intended to enhance linkage to opioid use disorder treatment, including medications for treating opioid use disorder, were included when they were retrieved by the literature search, although the search strategy was not engineered to fully capture all relevant reports in this area. For a more in-depth review of models of care for medication-assisted treatment for opioid use disorder, see Korthuis et al. (2017). Evaluations of strategies to address other opioid-related harms (e.g., HIV/HCV) that also aim to prevent overdose (e.g., comprehensive syringe services programs; see Des Jarlais et al., 2015), were not specifically searched for inclusion in the review. Drug consumption/safe injection facilities (e.g., see Andresen and Boyd, 2010) which are available internationally but not Federally sanctioned in the United States were excluded. Interventions implemented in community settings were included; interventions focused within correctional settings (e.g., enhancing access to medications for opioid use disorder in jails or prisons) were excluded.

The database search identified a total of 12,558 unique publications (Fig. 1). After screening titles, 1246 publications that clearly did not meet screening criteria were removed. Of the remaining 11,312 publications, we retained 156 publications for inclusion in the review based on abstract review; an additional 23 publications were nominated by study authors based on reference list searches and general awareness of reports (N = 179 total new publications). When questions arose about inclusion criteria, selection was determined after discussion and agreement among authors. The 179 new publications combined with 72 publications retained from the original systematic review resulted in a total of 251 publications included in the review.

The categories of interventions identified include: 1) state prescribing legislation or regulation, 2) prescription drug monitoring programs (PDMPs), 3) insurance and pharmacy benefit management strategies, 4) clinical guideline implementation, 5) provider education, 6) clinical health system interventions, 7) naloxone education and distribution, 8) safe storage and disposal, 9) public education, 10) community coalitions, and 11) interventions employing public safety and public health collaboration. Interventions covered in the current review differ slightly from the previous review (Haegerich et al., 2014) given the evolving nature of the research in this area and the addition of innovative approaches evaluated in the field. Some intervention categories remained similar in scope (e.g., PDMPs), some categories were further separated for clarity (e.g., guideline implementation, provider education, and health system interventions are now distinguished categories), and other categories were newly added (e.g., public safety and public health collaboration).

The authors independently reviewed articles within assigned intervention categories, documenting the study design and outcomes examined (see Supplementary Tables 1–11). The study design algorithm from the CDC Community Guide was used to categorize study design (Zaza et al., 2000). When questions arose about study design or outcomes, one author consulted a second author for discussion and validation until agreement was reached. Given the variation in interventions, study designs, and outcomes assessed, it was not practical to synthesize the results through meta-analysis. We developed a narrative synthesis for each intervention area, summarizing results from studies included in the previous review along with new studies to provide a holistic picture of the available evidence, focusing on the studies with the greatest scientific rigor in the narrative review (e.g., use of randomized trials, time series analysis, and comparison groups).

Authors assigned evidence quality ratings for each intervention category, including very low, low, moderate, or high, according to methods used in the previous review (see Haegerich et al., 2014) and inspired by the GRADE approach (Balshem et al., 2011). In brief, intervention categories comprised of primarily observational studies would receive an initial assignment of low evidence quality, while categories comprised of primarily randomized controlled trials would receive an initial assignment of high evidence quality. Ratings could be modified downward based on factors such as study limitations, inconsistency of results, and indirectness of evidence; ratings could be modified upward based on factors such as large magnitude of effect and dose response. When quality of evidence is low, there is low confidence that the effect is a true effect, and further research is likely to change judgments of effectiveness. When quality of evidence is high, there is confidence that the true effect is close to what evaluations have estimated. Studies were considered to offer indirect evidence when a health outcome (e.g., opioid overdose) was not assessed.

State policy (e.g., legislation and regulations) intended to prevent opioid overdose were examined in 14 studies (see Table 1, Supplementary Appendix), including policies addressing pain clinics (often in combination with PDMPs), immunity from prosecution, physical examinations, and comprehensive regulation. Because multiple policies are often implemented simultaneously within and between states, there are significant challenges in assessing the unique impact of specific state policies. The most rigorous studies employing time series designs with concurrent comparison states have evaluated pain clinic legislation; particularly, in Florida, and in combination with implementation of Prescription Drug Monitoring Programs (PDMPs). Among the most rigorous studies in Florida, two found a significant decrease in opioid volume and morphine milligram equivalent (MME) prescribed, especially among higher-risk prescribers (Chang et al. (2016); Rutkow et al. (2015)). In a study comparing opioid overdose mortality rates in Florida to North Carolina, a state without pain clinic regulations, Kennedy-Hendricks et al. (2016) estimated that Florida’s legislation saved 1029 lives in the 34 months following implementation in the state. Other, less rigorous studies in Texas have also found decreases in the amounts of opioids prescribed after implementation of pain clinic regulation, although in one study decreases were detected only in the short-term (Lyapustina et al. (2016); Raji et al. (2017)).

Some have raised concerns that regulating pain clinics could result in a shift in patients receiving opioids from clinics to obtaining opioids on the illicit market, and thus increasing overdoses from illicit opioids. In a time series analysis (without comparison), Johnson et al. (2014) reported a 27% decrease in the prescription opioid overdose death rate after policy changes in Florida. There was a concurrent increase in the heroin overdose death rate, but this increase was relatively small (from 0.3 to 0.6 per 100,000) and it is unclear whether this increase was due to state policy changes or due to changes in the illicit drug supply and secular drug use patterns during the study period. A national study conducted by Dowell et al. (2016) found that the combined implementation of pain clinic policies and mandatory PDMP review was associated with a significant 8% decrease in opioid prescribing and 12% decrease in prescription opioid overdose deaths (pain clinic policies alone did not significantly reduce these rates); however, no significant increase was detected for heroin overdose rates.

There is very limited evidence of the impact of Good Samaritan laws – laws providing varying degrees of legal protection for individuals engaged in seeking help during an overdose event. A cross sectional analysis of overdose mortality in all states showed that Good Samaritan laws were associated with a significantly lower (15%) incidence of overdose mortality (McClellan et al., 2018) while a non-comparative study reported an increased willingness to call 911 among people using opioids once the law was understood (Banta-Green et al., 2011).

We assigned the overall quality of evidence as low for state policies overall given the limitations of the study designs. However, evidence supporting the combination of pain clinic and PDMP legislation could be upgraded to moderate given enhanced rigor of study designs and consistency in findings.

State prescription drug monitoring programs (PDMPs) were evaluated as the primary intervention in 26 studies (Tables 1 and 2, Supplementary Appendix). Eight studies evaluated PDMPs in combination with other state policies [Al Achkar et al., 2018; Chang et al., 2016; Dowell et al., 2016; Johnson et al., 2014; Meara et al., 2016; Penm et al., 2017; Rutkow et al., 2015; Surratt et al., 2014]). In the studies evaluating PDMPs as the primary intervention, designs were structured to either test their overall impact once established, or to assess impact of specific PDMP characteristics.

Because PDMPs have evolved significantly since their inception, the most recent analyses have focused on PDMP characteristics that are intended to increase their utilization (e.g., mandatory registration and/or use), and the degree to which such characteristics influence prescribing and health outcomes. In the most rigorous studies, compared to states without mandatory use, PDMPs with mandatory use demonstrated significant decreases in the days’ supply of opioids prescribed (Yarbrough, 2018), as well as deaths involving prescription opioids when combined with pain clinic regulation (Dowell et al. (2016). More broadly, stronger PDMP states, such as those that required mandatory use, monitored more than schedule II drugs, and updated more frequently (e.g., daily), demonstrated greater reductions in overdose deaths involving prescription opioids (Pardo, 2016).

However, findings of impact on prescribing and health outcomes are mixed in the most rigorous studies examining the impact of the presence of PDMPs alone. We identified 10 studies that implemented time series analyses examining PDMP states with concurrent comparison states. Of the six studies that examined impacts of presence of a PDMP on prescribing behavior, four found significant decreases in prescribing, as evidenced by prescriptions for schedule II opioids (Bao et al., 2016; Simeone and Holland, 2006), oxycodone shipments, (Reisman et al., 2009), and opioid volume among Medicare enrollees (Moyo et al., 2017); however, two studies were unable to detect significant differences in average MME dispensed in states with PDMPs compared to those without (Brady et al., 2014; Paulozzi et al., 2011). Of the three studies that examined impact on overdose, two found no significant changes or differences in drug or opioid overdose mortality (Nam et al., 2017; Paulozzi et al., 2011). Yet, one found significantly lower opioid-related death rates in states with a PDMP compared to those without, particularly when the PDMP was more robust in terms of number of drug schedules monitored, mandated use, and update frequency (Patrick et al., 2016); estimating there could have been 600 fewer opioid overdose deaths in 2016 if Missouri adopted a PDMP and other states enhanced their programs. In two studies examining treatment admissions in PDMP states compared to non-PDMP states, one study found a significant decrease in PDMP states (Simeone and Holland, 2006) while the other did not (Reifler et al., 2012).

Findings are likely mixed due to the complexities involved in isolating the impact of PDMPs within the context of a worsening epidemic, a rapidly shifting intervention landscape, varying PDMP characteristics across states, and lack of information about specific implementation strategies (e.g., provider education on use). Yet, we assigned the level of evidence as moderate, particularly when examining impact on prescribing behavior, when taking into consideration both the study designs employed and the consensus of findings. Additional rigorous examinations of the impact of PDMPs on distal health outcomes are necessary, including more fine-tuned examination of PDMP characteristics that could influence effectiveness (e.g.. mandatory use and registration, delegate access, reporting frequency, and inclusion of risk scores to identify high-risk patients). Studies are also needed to examine actual provider PDMP use, beyond overall policy implementation.

Insurance interventions, such as those that identify high-risk opioid prescribing, lock in patients to specific providers or pharmacies (i.e., patient review and restriction), or require prior approval of medications before reimbursement, were evaluated in 22 studies (Table 3, Supplementary Appendix). Approximately half of the studies utilized a more rigorous time series design (N = 10), and one employed a randomized controlled trial design.

Three time series studies evaluated lock-in programs within Medicaid populations. All reported positive changes in prescribing behavior, including significant decreases in polypharmacy, increases in number of patients filling opioids only from assigned prescribers, and decreases in quantities of opioids prescribed or filled (Blake et al., 1999; Mitchell, 2009; Skinner et al., 2016). Mitchell (2009) reported a significant decrease in emergency department visits as well (the other more rigorous studies did not examine health outcomes). However, less methodologically rigorous studies of lock-in programs reported an unintended consequence: a significant increase in non-Medicaid reimbursed opioid prescriptions (e.g., cash payment) after program implementation (Naumann et al., 2018; Roberts et al., 2016).

Three time series studies focused on prior authorization (PA) policies in the Medicaid population. Morden et al. (2008) found that only strict PA policies – such as those implementing “fail first” protocols and requiring medical documentation for opioid use for pain – were associated with a significant decrease (34%) in oxycodone use. Another study found that PA policies for extended release/long acting (ER/LA) opioids significantly decreased prescriptions for those types of medications, including high-dose prescriptions (Oregon State University (OSU, 2012). Such prescribing outcomes could be considered to be positive impacts. However, a lower quality study found that a decline in ER/LA opioid prescribing after prior authorization coincided with a significant increase in prescriptions for short acting opioids illustrating a displacement effect; further, no differences in ED visits and hospitalizations were seen, questioning whether changes in prescribing translate to changes in health outcomes (Keast et al., 2018). In evaluating a PA policy for buprenorphine used in medication treatment for opioid use disorder to limit dose and prescription length, Clark et al. (2014) noted a decrease from 16.5% to 4.1% in doses exceeding 24 mg per day (the maximum recommended therapeutic dose), accompanied by a temporary increase in relapse for some patients. Hence, while prior authorization for prescription opioids might be helpful in terms of changing prescribing behavior, with uncertain impacts on health outcomes, prior authorization for treatments for opioid use disorder have the potential for harm by facilitating resumption of drug use.

Pertaining to drug utilization review, two methodologically rigorous studies using time series and randomized trial designs evaluated retrospective drug reviews of patients by employing letters mailed to providers that prescribed to high-risk patients. Both studies illustrated significant decreases in the number of controlled substances filled, with one illustrating greater reductions when letters provided more patient-specific information (Daubresse et al., 2014; Gonzalez and Kolbasovsky, 2012). Similarly, Zarowitz et al. (2005) reported a decrease in both the number of prescriptions per member and in polypharmacy events after the implementation of a pharmacist-led provider education intervention based on the results of drug-utilization review. Unfortunately, no data are available from rigorous studies about impacts of drug utilization review on patient health outcomes.

In an evaluation of a comprehensive commercial payer opioid utilization policy that included patient-provider agreements, risk assessments, lock-in programs, PA policy, a mail-order ban, and quantity limits, there were significant decreases in both the short-acting opioid prescription rate (−6.1%) and the LA opioid prescription rate (−9.1%) (Garcia et al., 2016). Similarly, Garcia et al. (2014) evaluated a comprehensive Medicaid initiative aimed at reducing LA opioid prescribing and found a significant, 17.8% reduction in the number of members utilizing LA opioids.

Reviewed studies indicate decreases in prescribing are possible through requiring prior authorization for opioids for pain management, restricting the number of providers and pharmacies that can prescribe and dispense to a patient when patients are identified as high-risk, and reviewing medication use with patient-specific provider feedback; comprehensive policies hold particular promise. We assigned an evidence strength of moderate for prescribing behaviors, but more research is needed to evaluate the impact on health outcomes such as overdose, as well as improvements in patient pain and function, such as through encouraging use of non-opioid strategies. Concern has been raised that insurance policies motivated by clinical guidelines but implemented inflexibly or without adequate consideration of patient context could result in patient harm; thus, further evaluation of insurance strategies are warranted to assess for intended as well as unintended consequences (Dowell et al., 2019; Kroenke et al., 2019). Additional research on prior authorization policies focused on medications used for treatment of opioid use disorder could inform how such policies might positively or negatively affect treatment access, retention, relapse, diversion, overdose, and other outcomes.

National, state, local, and professional society clinical practice guideline implementation was evaluated in 34 studies (Table 4, Supplementary Appendix). Evaluated guidelines focused primarily on opioid prescribing, such as limiting the dosage or duration of prescriptions, using PDMPs and urine drug testing, starting with short-acting opioids, and avoiding co-prescribing of opioids and benzodiazepines. Strategies for implementation included limited efforts of distribution (e.g., paper/ electronic dissemination of recommendations, small group training), as well as intensive efforts such as integration of recommendations within the electronic health record (EHR) that provides clinical decision support at the point of care, and structured educational visits to offer tailored training in evidence-based practice (i.e., academic detailing). Several studies (n = 10) examined impact of opioid prescribing guidelines on provider behavior using before/after designs, with only a minority of studies (n = 7) examining patient health outcomes. State guidelines were evaluated using more rigorous time series designs (n = 7). Randomized trials (n = 5) were employed primarily to test different ways of presenting guideline information and improving knowledge, without “no treatment” controls.

At a national level, one rigorous interrupted time series analysis of the 2016 CDC Guideline for Prescribing Opioids for Chronic Pain illustrated significant declines in the overall opioid prescribing rate after guideline release, beyond existing declining trends, including decreases in the rate of high-dosage opioid prescriptions and overlapping opioid and benzodiazepine prescriptions (Bohnert et al., 2018). Time series analyses of the Washington State opioid dosing guideline have illustrated declines in number and amount of opioid prescriptions, as well as fewer patients newly initiating opioids becoming long-term users in a worker’s compensation population (Franklin et al., 2012; Garg et al., 2013). Significant changes in prescribing were also detected among the Washington state Medicaid population, with significant decreases in high MME dosages and use of LA opioids overall, and for dispensing after an ED visit specifically, including among patients with prior risky opioid use (Sullivan et al., 2016; Sun et al., 2017). Similarly, reductions in opioid prescriptions dispensed through EDs were seen after Ohio emergency physician opioid prescribing guidelines were released, compared to pre-guideline trends; impact on patient health outcomes were not assessed (Weiner et al., 2017). When patient health outcomes have been assessed, findings are mixed. In the Washington state worker’s compensation population after guideline implementation, one study illustrated a decrease in number of unintentional deaths related to prescription opioid use (Franklin et al., 2012) and another study found no declines in ED visits for poisoning or adverse effects involving opioids (Fulton-Kehoe et al., 2013). Yet, this was during a time of rapid increases in ED visits nationally.

Guideline impact can be influenced by implementation strategies. For example, communication strategies that provide recommendations within the clinical context (i.e., present information in narrative format rather than summary format) can enhance knowledge improvements (Kilaru et al., 2014; Meisel et al., 2016). To achieve behavior change, however, additional efforts may be required. In rigorous randomized controlled trials, investigators have found that education, monitoring, and feedback to providers on guideline implementation for management of chronic pain are sometimes not sufficient to facilitate use of recommended practices or improvements in patient outcomes (Corson et al., 2011; McCracken et al., 2012). However, pairing guidelines with more intensive efforts, such as academic detailing, systems consultation, and coordinated care plans illustrates particular promise for changing prescribing behavior (e.g., decrease in high opioid dosages) and influencing patient outcomes, such as decreased overdose mortality (Paone et al., 2015; Quanbeck et al., 2018; Von Korff et al., 2016).

We determine the overall quality of evidence to be low given the limitations of designs in many studies (e.g., pre-post only without control), mixed findings on behavioral outcomes, and the limited number of studies examining patient health outcomes. However, there were several studies with strong designs, such as time series designs, that illustrated positive impacts on prescribing behaviors. More intensive implementation strategies, such as making recommendations available at the point of care and using academic detailing and coordinated care plans implemented in tandem with guideline dissemination, hold particular promise for changing prescribing behavior and decreasing overdose mortality. Similar to considerations for insurance strategies, further research is needed to identify impacts on patient health outcomes including both intended and unintended consequences, given concerns about harms associated with implementation of recommended practices without adequate consideration of patient characteristics (Dowell et al., 2019; Kroenke et al., 2019).

Provider education was evaluated in 32 studies (Table 5, Supplementary Appendix). The predominance of studies (n = 24) evaluated provider education on opioid prescribing for acute or chronic pain; four studies included education on opioid use disorder and prescribing medication for opioid use disorder treatment, and six included education on overdose and prescribing or dispensing naloxone. Some educational programs focused on students or residents, while others focused on educating experienced providers, with some targeted to providers based on a history of high risk prescribing (e.g., referral by the medical board). In some cases, providers were educated about their prescribing history specifically, in comparison to other providers. Educational strategies included in-person presentations and workshops, web-based training, presentation combined with direct application in the clinical setting (e.g., in combination with structured clinical examination), and telehealth strategies with case-based learning (e.g., project ECHO). Most studies were descriptive, with only 2 studies using randomized trial designs and 9 studies using comparison groups. Most studies evaluated the impact of education on knowledge improvements, self-efficacy, and self-reported behavioral intentions or changes in practice; only one third of studies evaluated direct changes in prescribing behavior or other objective changes in provider behavior (e.g., PDMP registration, use of risk mitigation strategies), and none evaluated impacts on patient health outcomes.

In one randomized trial, Michael et al. (2018) offered emergency department physicians feedback on how their prescribing behaviors compared to their peers, finding that those in the intervention group who had initially underestimated their own prescribing had a significantly larger decrease in prescribing than physicians in the control group. In another randomized trial, residents who received web-based training on guideline recommendations had a greater increase in knowledge and self-rated competence of patient management than residents who only received the guideline document (Sullivan et al., 2010). Other findings from before and after studies with comparison groups revealed that provider education generally is effective in increasing provider knowledge and self-efficacy, intention to change, and self-reported change in prescribing behavior (e.g., Alford et al., 2016; Cardarelli et al., 2018; Zisblatt et al., 2017). One study found no differences in knowledge between in-person versus online training format (Berland et al., 2017), with another finding that online training out-performed document dissemination (Kim et al., 2016). In terms of objective measures of provider behavior, Ury et al. (2002) illustrated improvements in recommended prescribing practices after education on palliative care and pain management. Telehealth education, namely within the Project ECHO model, increased provider knowledge and self-efficacy for opioid prescribing for pain, with some evidence for change in prescribing behavior, referrals, and use of risk mitigation strategies (Anderson et al., 2017). Mixed findings have been found after naloxone training for providers, with some programs showing no differences in knowledge (Jacobson et al., 2018), and others showing increases in naloxone prescribing rates (Taylor et al., 2018).

We assigned an overall evidence quality of low given the limitations of the study designs (e.g., primarily before-after studies without true control), mixed findings on objectively measured provider behavior outcomes, and the lack of studies examining patient health outcomes. Provider education that goes beyond curriculum presentation with additional supports (e.g., provider feedback or tools) could increase education effectiveness. Distance education through telehealth can increase knowledge and practice changes for providers who have difficulty accessing in-person education opportunities. More rigorous research is needed to identify if education that focuses on obtaining waivers for prescribing buprenorphine for opioid use disorder has the potential to increase capacity for treatment, in particular.

Clinical health system interventions were evaluated in 33 studies (Table 6, Supplementary Appendix). A wide variety of interventions are included in this category, including: integrated pharmacist services and multidisciplinary team approaches to manage pain; brief motivational interviewing to reduce opioid misuse or overdose risk; patient education on pain management, opioid risks, and overdose; EHR clinical decision support and opioid quantity defaults; audit and feedback of providers on opioid prescribing; and system policies concerning opioid dosage and risk mitigation strategies. Compared to other categories of interventions evaluated in this review that are most directly relevant to clinical care (e.g., guideline implementation, provider education), evaluations of health systems incorporated more rigorous study designs, including randomized trials (n = 14) and time series analyses (n = 3); however, descriptive and before/after studies were also numerous. Studies primarily assessed impacts on provider and patient behavior; three studies examined patient health outcomes.

Among randomized or nonrandomized trials, multidisciplinary team approaches have been shown to reduce opioid prescribing after intervention (Neven et al., 2016). Patient education about opioid risks and overdose can increase patient knowledge and behavioral intentions (Dunn et al., 2017; McCarthy et al., 2015). Mehl-Madrona et al. (2016) demonstrated that patients educated on nonpharmacologic pain treatment who remain engaged in support sessions significantly reduced or discontinued opioid use, with reductions in pain and improvement in quality of life.

Compared to patients receiving usual care, patients engaged in brief interventions by clinical providers, including motivational interviewing about their opioid misuse, report less misuse after intervention (Bernstein et al., 2005; Bohnert et al., 2016; Gelberg et al., 2015; Gryczynski et al., 2015). Mobile psychosocial interventions for increasing adherence to medications for opioid use disorder treatment also shows promise (Guarino et al., 2016). However, findings are mixed when more objective measures are assessed (e.g., through toxicology testing), and in some cases, significant differences in outcomes have not been detected between intervention and control groups (Ondersma et al., 2014; Saitz et al., 2014) or between patients receiving in-person consultation versus computer administration (Schwartz et al., 2014). In one study, motivational interviewing unintentionally reduced the likelihood of receipt of substance use disorder treatment (Kim et al., 2016). Yet optimistically, in a pilot randomized controlled trial, Coffin et al. (2017) demonstrated that individuals with opioid use disorder and overdose history who engaged in motivational interviewing including follow-up counseling in a public health naloxone education and distribution program experienced a significant decrease in overdose events compared to individuals engaged in overdose education as usual.

Providing decision support at the point of care, such as via the EHR, and providing feedback to providers on their prescribing was found to decrease opioid prescribing after implementation (Gugelmann et al., 2013; Lin et al., 2017) as well as to decrease ED visits by patients (Ringwalt et al., 2015). System policies such as opioid dosing limits in combination with provider education have resulted in reductions in opioid prescribing and dosage, with no significant differences in pain or quality of life among patients after policy initiation (Weimer et al., 2016).

We judge the overall quality of evidence to be moderate. Study designs included randomized trials and time series analyses, and studies have provided overall consistent findings of improvements in provider and patient behavior, and in some cases patient health outcomes. Randomized studies did include small sample sizes, however, and other important limitations such as high attrition. Findings suggest that educating patients can improve knowledge and intentions, but interventions such as motivational interviewing that engage patients in understanding discrepancies in behavior, resolving ambivalence, and exploring motivations and plans for change hold particular promise in reducing risky behavior and affecting health outcomes among those at highest risk for overdose. Multidisciplinary interventions and those that engage providers on recommended clinical practices at the point of care with feedback on prescribing behavior have the potential to change not only provider behavior but patient health outcomes as well.

A total of 65 studies examined naloxone education and distribution (Table 7, Supplementary Appendix). The venue for naloxone education and distribution varied and included naloxone co-prescribing with opioids, first responder training, pharmacy-, health system-, and community-based naloxone distribution, and bystander engagement.

Patient behaviors, such as acceptance of a naloxone kit, were the outcomes of interest for the majority of studies (n = 33). Fewer studies examined provider behavior (n = 27) or health outcomes such as number of reversals (n = 20). Most studies within this section used low-quality study designs, such as non-comparison (n = 34) or before-and-after (n = 18); however, three randomized controlled trials evaluated impacts on patient behavior and health outcomes.

The preponderance of studies employed descriptive techniques that illustrated characteristics of naloxone programs, policies, and distribution sites; the types of participants receiving services; the number of participants trained; or amount of naloxone distributed. In prospective cohort and randomized trials, patients receiving naloxone training and kits exhibited significantly greater knowledge of overdose symptoms and when naloxone was indicated after training (Green et al., 2008; Huhn et al., 2018). Reported use of distributed naloxone kits in reversing an overdose has also been documented (e.g., Papp and Schrock, 2017; Spelman et al., 2017; Walley et al., 2013b).

At the community level, targeted messaging to increase public support has been found to be most effective in increasing support for naloxone policies when it includes both factual information and a sympathetic narrative (Bachhuber et al., 2015). A time series analysis with concurrent controls identified that overdose death rates were significantly reduced in communities with opioid education and naloxone distribution (OEND) programs compared to communities without these programs (Walley et al., 2013a). Further, a time series analysis of data from a parent randomized control study found that individuals involved in syringe services programs with naloxone education and distribution reported fewer days of heroin use compared to baseline use (Kidorf et al., 2013).

In the clinical environment, a retrospective cohort study found that providers that had at least one academic detailing session were more likely to prescribe naloxone to their patients (Bounthavong et al., 2017). However, another retrospective cohort study showed no significant difference in naloxone administration rates between patients who received naloxone education and receipt than those that did not (Doe-Simkins et al., 2014). In a nonrandomized intervention study, Coffin et al. (2016) documented a decrease in opioid-related ED visits after providers and clinic staff were trained in naloxone prescribing, with a focus on indications for prescribing, language to use with patients, formulations, payer coverage, and naloxone use. However, in a randomized trial, Banta-Green et al. (2011) conducted overdose education, brief counseling, and naloxone prescription for patients at elevated risk for an overdose after an ED visit and found that overdose events did not significantly differ between intervention and control participants.

We determined the quality of evidence to be low given study designs, despite the preponderance of evidence of naloxone as a vital clinical tool and consensus of the large volume of findings. Studies with more robust methodological designs could identify the most effective training and delivery modalities and for which populations. Which other components may be necessary for effective harm reduction (e.g. wrap-around care for infectious diseases, such as comprehensive syringe services programs and linkage to treatment after an overdose) should be explored.

Safe storage and disposal interventions were evaluated in 13 studies (Table 8, Supplementary Appendix). Studies were predominantly non-comparative (N = 6) or before-after designs (n = 3) focusing on patient behaviors. Two cross sectional studies and one before-after study focused on wide-reaching media campaigns that included safe storage and disposal components, examining target population reach, attitudes, and self-reported behaviors. Non-comparative descriptive studies primarily reported on amounts and proportions of discarded drugs across different geographic areas. Findings generally showed small percentages of controlled substances, including opioids, disposed relative to disposal of other types of medications.

In a before-after study with a concurrent comparison group, Hasak et al. (2018) educated surgery patients on the safe disposal of opioids with an educational brochure and found a doubling of the disposal rate (11%–22%) after brochure distribution. De La Cruz et al. (2017) evaluated a patient educational program in a cancer outpatient palliative care clinic that uses educational material focused on proper prescription opioid use, storage, and disposal in a non-randomized trial. After receiving the intervention, patients had significantly greater knowledge of proper disposal methods (76% vs. 28%) and were significantly less likely to practice unsafe use of prescription opioids (18% vs. 25%) compared to the control group. In the only RCT identified, a program promoting pharmacy-based drug disposal after dental surgery was evaluated. Compared to the control group, patients who received the brief behavioral intervention were 22% more likely to report disposal or intent to dispose of unused prescription opioids; however, the sample size was small and results were not statistically significant (Maughan et al., 2016).

Safe storage and disposal interventions could play an important a role in limiting the availability of prescription opioids for diversion, but we judge the quality of the available evidence as low. Safe storage and disposal education through both multimedia campaigns and patient-targeted material can increase participant knowledge, but little is known about how that translates to health outcomes, such as reductions in misuse or overdose. Rigorous studies evaluating the impact of disposal boxes and take-back events on health outcomes are needed. Most studies have limited sample sizes with before-after designs that focus on patient behaviors. Other study limitations include the lack of baseline data and comparison groups, unassessed health outcomes, and other events occurring simultaneously that could be responsible for effects. Although findings are encouraging, they mainly highlight the need for rigorous evaluations examining health outcomes.

Public education interventions were evaluated in 8 studies (Table 9, Supplementary Appendix; with one study also reviewed for safe storage and disposal). Given the small number of studies and variety of content and modalities used, findings are reviewed from all studies, regardless of methods employed. Studies included a mix of before/after designs (n = 2) and small randomized trials (n = 6), with a predominant focus on patient behavior outcomes. In the clinical setting, Chakravarthy et al. (2018) illustrated that video-based education on safe opioid use upon discharge in the ED can improve patients’ recall about safe opioid use and disposal. Similarly, Guarino et al. (2018) documented in a randomized trial that a web-based behavioral intervention for patients with chronic pain and a history of opioid misuse reduced aberrant drug-related behavior and ED visits compared to patients receiving treatment as usual. Pre-operative education on safe opioid use was demonstrated in a randomized trial to encourage discontinuation of opioid use, and lower opioid consumption six weeks and three months after surgery (Syed et al., 2018).

Klisch et al. (2013) examined the effectiveness of forensic science games that address drug misuse for integration in the school science curricula; compared to pretest, student attitudes were more negative toward drug abuse after playing the games. In a small study, Fang and Schinke (2013) demonstrated that an online substance abuse prevention program for mothers and daughters that focused on relationships and resilience reduced intentions and self-report of substance misuse. The most rigorous evidence comes from evaluations of universal school and family-based preventive interventions focused on developing youth competencies (Crowley et al., 2014; Spoth et al., 2013). Such programs focus on nurturing skills that support children and dealing with peer pressure for parents, and self-management, social skills, and drug resistance skills for children, and have evidenced significant reductions in prescription opioid misuse. Finally, a before/after evaluation of the “Use Only As Directed” statewide media campaign in Utah that included messages about safe use of prescription pain medications and proper disposal through TV/radio spots, posters, advertising, news releases, and information cards found that survey respondents were less likely to share medications and take medications not prescribed to them after the campaign. Implementation was accompanied by a 14% one-year reduction in unintentional opioid-related drug overdose deaths (Johnson et al., 2011).

Overall, education efforts have some, yet limited, promise for increasing knowledge and awareness of opioid-related harms and decreasing substance misuse, with the most rigorous evidence coming from intensive school and family-based interventions for youth and randomized trials in the clinical setting focused on patient use of opioids; impact on health outcomes was not assessed. Overall quality of evidence for this heterogeneous group of interventions is judged as moderate, given the strength of study designs, and assessment of misuse (in the case of school and family-based interventions). Media campaigns to prevent opioid overdose have promise, yet require further rigorous evaluation, particularly given lack of evidence for changes in health outcomes, and mixed findings from previous evaluations of campaigns focused on illicit drugs beyond opioids, not included in the current review due to their expanded scope (Allara et al., 2015).

Interventions involving coalitions formed to coordinate opioid overdose prevention efforts across a state were evaluated in 3 studies (Table 10, Supplementary Appendix). All three studies evaluated Project Lazarus in North Carolina. Project Lazarus includes coordination among county health departments, chronic pain initiatives, and substance abuse task forces on overdose prevention. Albert et al. (2011) and Brason et al. (2013) reported reductions in overdose mortality, but impacts of Project Lazarus independent of other state- or local-level interventions could not be determined. In a time series design with concurrent comparison groups, Alexandridis et al. (2017) documented in the longer-term, programs for patients with pain were associated with lower mortality rates, and naloxone policies and expansion of medication-assisted treatment was associated with lower ED visits (but the effect of treatment stabilized and rates began to increase again). Diversion control was associated with counter-intuitive increases in ED visits.

Interventions involving coalitions could play an important role in the coordination of opioid overdose prevention efforts but we judged the quality of the evidence to be very low with more research needed. More rigorous evaluations are needed to analyze the independent effects of such coalitions on health outcomes of participants.

Interventions involving collaborations between public safety and public health in the community were assessed in 2 before-after studies evaluating policy-led linkage to treatment, both of which are reviewed here (Table 11, Supplementary Appendix). Medications for opioid use disorder treatment is recognized as the most-effective treatment for opioid use disorder (SAMHSA, 2018). However, medication-based treatment is underused and linkage to treatment could help improve treatment access and health outcomes for individuals recently treated for an overdose.

The first study, a training program on opioid overdose prevention and naloxone administration for law enforcement officers with a referral to treatment component resulted in approximately 20% of patients seeking treatment after an overdose (Dahlem et al., 2017). Officers were also significantly more knowledgeable about naloxone and drug overdose reversal after the intervention. The second study, the addiction treatment referral program led by the Gloucester Massachusetts police department was successful at accommodating 75% of participants who sought a referral (Schiff et al., 2017). Although the difference was not statistically significant, program participants were more likely to report drug abstinence at the mean follow-up time of 6.7 months compared to individuals who did not participate (40.9% v. 26.0%). Although these two studies lack rigorous designs and large sample sizes, the results show some promise.

Overall, we assess that the quality of evidence supporting public safety and public health collaborations to prevent opioid overdose is very low. More rigorous evaluations of public safety-led linkage to treatment interventions in the community are needed to determine their effectiveness in improving access to medication-based treat.ment and long-term health outcomes. Additional studies could address the effectiveness of strategic use of surveillance data to inform public safety and public health partnership efforts. For example, collaborative efforts such as the RxStat model established in New York City that aim to share health and public safety data among stakeholders to target and tailor programs, practices, and policies to address opioid misuse and overdose may hold promise (Heller et al., 2014). Further, efforts to improve the provision of medication-based treatment during incarceration and enhance linkage to treatment and ongoing supports upon release (beyond the scope of this review) have demonstrated positive impacts on overdose, but could benefit from further evaluation (Green et al., 2018).