Association between supportive interventions and healthcare utilization and outcomes in patients on long-term prescribed opioid therapy presenting to acute healthcare settings: a systematic review and meta-analysis

The opioid epidemic is a major public health problem across the world. While not initially recognized as a crisis, it has become a public health emergency that is largely believed to have begun in 1996 when OxyContin® was approved by the FDA [1]. Opioid prescription was initially thought to be the crux of the problem and multiple interventions were implemented in response over the years. Most of these interventions were focused on reducing access to opioid prescriptions with unforeseen and paradoxical consequences. Indeed, strictly targeting a reduction in opioid access ignored the complex socio-economic factors that impact opioid-related morbidity and mortality. As an example, reductions in prescription rates in Ontario (Canada) lead to a doubling of opioid mortality due to an increase in street opioids [2]. This shift in the nature of the opioid crisis to illicit opioid use has indeed been seen in multiple locations over the years and paralleled these interventions [3]. Considering such findings, illicit drug use and harm reduction related to their use has become the focus of most current efforts to limit the impact of the opioid crisis and the impact of opioid prescriptions has taken a backseat. However, it remains that there are patients on long-term opioids for chronic painful conditions without diagnosed or suspected opioid use disorder that are also at risk for complications. This group has ultimately been somewhat neglected amidst the extensive literature available on the more obvious high-risk populations. It is however quite clear that individuals on long-term opioid therapy are at risk of poor outcomes, including hospitalization, overdose and death from the use or management of their opioids [4, 5]. These poor outcomes invariably lead to acute healthcare presentations that have significant system level impacts on health services use including emergency department (ED) presentations, hospital and intensive care unit (ICU) admissions, as well important socio-economic consequences [6‐10].

Harm reduction strategies are defined as “any policy or program designed to reduce drug-related harm without requiring the cessation of drug use; these interventions may be targeted at the individual, the family, community or society [11‐19].” Harm reduction strategies are traditionally thought of and applied to patients demonstrating high risk behaviors and the term has not typically been applied in relation to individuals on long-term opioid therapy without a diagnosed or suspected opioid use disorder. These individuals may still benefit from harm reduction strategies as they require these medications for their well-being but may still suffer harmful consequences from their use. Harm reduction strategies are however traditionally associated with opioid use disorder. There is thus ample evidence for harm reduction strategies in acute opioid overdoses that is lacking for individuals on long-term opioid therapy without opioid use disorder. Their impact may be substantial given the reported high risk of adverse events in these settings [7, 20‐22]. As the strategies applying to this population may be different, we will instead refer to them as “supportive interventions” to limit confusion with strategies associated with opioid use disorder.

The most important complications as a result of opioid therapy most often lead to a presentation to an acute healthcare setting, defined as a setting in which “health system components, or care delivery platforms, are used to treat sudden, often unexpected, urgent or emergent episodes of injury and illness that can lead to death or disability without rapid intervention [23].” Examples of these include emergency departments, acute health care clinics, and hospital inpatient units. Accordingly, this may be the healthcare setting in which the most impactful supportive interventions may lie, and how this population can best be captured in studies. As such, the primary objective of this study was to identify the most effective supportive interventions for patients on long-term prescribed opioids presenting to acute care settings to decrease complications attributable to opioid use, to reduce avoidable health services use, and to improve outcomes.

We performed a systematic review using the guidelines from Cochrane and the Centre for Reviews and Dissemination [24, 25], and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guideline and the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines for observational studies [26, 27]. The study was registered with the PROSPERO (CRD42018088962 on 2018/02/19) International Prospective Register of Systematic Reviews (http://​ww.​crd.​york.​ac.​uk/​prospero).

The search strategy was developed and executed by an information specialist (RF) and was peer-reviewed by a second research librarian (Table 1). The information specialist searched electronic databases: Ovid MEDLINE (1946-), Ovid EMBASE (1996-), Cumulative Index to Nursing and Allied Health Literature (CINAHL) via EBSCOhost (1937-), and Wiley Cochrane Library (inception-), including the Cochrane Database of Systematic Reviews and the Cochrane Central Register of Controlled Trials (CENTRAL). We also searched reports from the National Information Center on Health Services Research and Health Care Technology (NICHSR) via the NICHSR ONESearch portal. Study records were also searched via the trial registry platform ClinicalTrials.​gov, guidelines via the National Guidelines Clearinghouse, and meeting abstracts via the Conference Proceedings Citation Index database (Clarivate Analytics). The following search themes was used: 1) opioids; 2) long-term drug therapy; and 3) acute healthcare settings (emergency departments, acute care surgery, critical care, urgent care and short-term inpatient stabilization). We additionally scanned the reference list of relevant included studies for additional articles. Bibliographic records were exported to an EndNote X9 (Clarivate Analytics, Philadelphia, Pennsylvania). Databases were searched up to and including April 11, 2019.

Articles were identified through two phases. First, two authors (JD1 and JG) independently reviewed the titles and abstracts of all retrieved articles for study inclusion. Second, full texts of the selected articles were retrieved, reviewed and selected based on inclusion criteria. All steps were performed in duplicate, and in each phase, disagreements were resolved through discussion. In the case of unresolved matters, a third author (OGR) was involved. Reasons for exclusion of full text articles were recorded and displayed in a PRISMA diagram format.

The methodological quality of each study was independently analyzed by two authors (JD1 and JG) using the Newcastle-Ottawa Quality Assessment Scale (NOS) for observational studies and the Cochrane Risk of Bias Tool for randomized controlled trials [24, 30]. Disagreements were resolved through discussion, and in the case of unresolved matters, a third author (OGR) was involved.

An inventory of supportive strategies was developed from the included studies. Descriptive analysis and data extraction of patient and study characteristics, and of supportive strategies and their outcomes were performed using standardized electronic data forms (Supplementary Appendix 1 for variables extracted). We analyzed all available data qualitatively and, when possible, aggregate analysis was performed [31]. For each study we extracted or computed the ratio of means between the intervention and usual care groups with 95% confidence interval [32]. These ratios were then pooled using the DerSimonian-Laird random effects method with an inverse variance weighting. To minimize bias due to confounding, results for RCTs were pooled separately from results for cohort studies and the two were not combined in one meta-analysis. Heterogeneity was assessed using the I-squared statistic with values greater than 50% considered to be “substantial” heterogeneity.

While this study had several important strengths (i.e., breadth of scope, rigorously pre-defined methodology stretching across several medical domains, presence of patient advisors), several important limitations warrant discussion. First, important terms (i.e., long-term medical opioid therapy, opioid ‘abuse’ and misuse, harm reduction strategies) were heterogeneously defined across studies and may have been a barrier to study identification. Most importantly, the supportive strategies were overall poorly defined across studies. Despite a careful analysis of the interventions to regroup or reclassify them under umbrella terms, it was difficult to clearly identify separate supportive strategies in some studies. Accordingly, these studies then often used multiple supportive strategies simultaneously, which significantly limited our ability to have a rigorous analysis. This is reflected in the meta-analysis where the most important harm reduction strategy was analyzed, acknowledging that it may not be fully separated from other minor elements of the intervention that may be classified under another umbrella term. We attempted to mitigate these factors by independent screening by two authors to ensure the inclusion of all relevant studies and appropriately classify the supportive strategies. Second, the rate of study inclusion was only 0.4%. This was secondary to most identified studies either studied illicit drug use or poorly differentiated long-term opioid use without opioid disorder from opioid use disorder. We aimed to exclude opioid use disorder or abuse but were faced with a high degree of uncertainty in some cases. We thus decided to include studies only if they specifically referred to acute pain presentation in long-term opioid users even if there was mention of prescription opioid misuse, as long as there was some clear distinction between the groups. Given that there is a degree of conversion from appropriate use to abuse, we believe that this captures well this evolution in patients. Similarly, identifying what constitutes harm or complications from long-term opioid use proved challenging. Presentations other than overdose or without the attached opioid misuse label are often unrecognized as related to opioids. We had to assume that in the selected papers, the focus on patients being on opioid therapy means that there is a reasonable expectation that their presentation is related to opioid in some way. In our opinion, acute on chronic pain qualified as such, as it either represents hyperalgesia or under-treatment, both of which require a measured treatment approach. Third, the wide scope of some supportive strategies lead to difficult decisions for study inclusion. Indeed, a number of harm reductions were part of a package organized at a state level. It was difficult to separate the specific impact of each strategy, the impact on acute versus non-acute healthcare settings and to discern which studies dealt with patients on appropriate long-term opioid therapy. In these situations, we opted to include these state level studies if there was a well-described significant proportion of long-term opioid users, and if number of acute healthcare presentations was an outcome of interest. We do acknowledge that these studies reflect a very heterogeneous group in a lot of instances and limit the validity of the findings. This is not reflected well in the quality assessment of the cohort study who are technically for the most part of moderate to high methodological quality. The RCTs are for their part paradoxically at moderate to high risk of bias due to their design but represent a more homogeneous population. Fourth, most identified studies were from the US, limiting the generalizability of our findings to other jurisdictions that may have different policies and context that affect the outcomes of the identified supportive strategies. This is not surprising as the opioid epidemic was first recognized in the US, and many findings in the US are applicable across Canada and other high-income countries [1]. Finally, while we decided to include studies from 1996, all of the studies included are from the last 15 years. This is likely explained by the delayed recognition of the public health crisis from the opioid epidemic.

Our systematic review revealed that most of the studies have targeted patients presenting to the ED, with very little data on inpatient supportive strategies. This knowledge gap is reflected in the most recent Canadian guidelines for opioid use for chronic non-cancer pain, which do not address acute admissions in this population [58]. These guidelines do reflect the importance of a multidisciplinary approach in the chronic non-cancer pain population, which would be similar to the “supports for patient in pain” harm reductions strategy. Studying this harm reduction strategy for non-ED acute healthcare settings would strengthen the current body of evidence. Importantly, studying these strategies using patient-related outcomes such as mortality, quality of life and pain is of paramount importance, as opioid prescriptions and ED visits appear to be poor or misleading surrogate endpoints. Future policy work informed by these results would lead to better resource utilization through a shift from reactionary processes (i.e., ED visits) to preventative strategies that prevent acute healthcare presentations.