Introduction

There is little consensus when it comes to defining polysubstance use (PSU); it is often used as a broad term to describe the use of ≥2 different substances or classes of substances either simultaneously or separately over a defined period (Connor, Gullo, White, & Kelly, 2014; Tomczyk, Isensee, & Hanewinkel, 2016). A growing body of international evidence suggests that PSU is particularly frequent among people who use opioids (PWUO) (Chen et al., 2019; Harrell, Mancha, Petras, Trenz, & Latimer, 2012; Hassan & Le Foll, 2019; Yang et al., 2018). Moreover, drug overdose deaths often involve several drugs in addition to opioids (Crummy, O'Neal, Baskin, & Ferguson, 2020). For example, in the Canadian province of British Columbia, the most detected drugs in illicit drug toxicity deaths during the past four years were fentanyl (83%), cocaine (50%), amphetamines (34%), and heroin (15%) (BC Coroners Service, 2020). Similar patterns are observed in the United States, where National Vital Statistics System reported cocaine overdose deaths involving any opioids increased from 29% to 63% between 2000 and 2015 (McCall Jones, Baldwin, & Compton, 2017).

Measuring PSU is statistically and methodologically challenging. Most studies have assessed PSU using contingency tables of several single drugs, leading to small cell sizes or highly skewed distributions and, therefore, limited or biased analyses or interpretations (Afshar et al., 2019; Connor et al., 2014; Crummy et al., 2020; Liu & Vivolo-Kantor, 2020; Liu, Williamson, Setlow, Cottler, & Knackstedt, 2018). Despite inconsistencies over what constitutes PSU, studies have consistently associated PSU with an increased risk of numerous poor health outcomes (Carter et al., 2013; Connor et al., 2013; Hedden et al., 2010; Quek et al., 2013; Timko, Han, Woodhead, Shelley, & Cucciare, 2018; Trenz et al., 2013).

Considering the importance of characterizing the distribution and correlates of PSU for overdose prevention programs and harm reduction interventions, methodological developments have been undertaken to help better identify and uncover subpopulations with distinct PSU patterns through person-centered methods (Lanza & Cooper, 2016; Lanza & Rhoades, 2013; Lanza, Collins, Lemmon, & Schafer, 2007; Tomczyk et al., 2016). In contrast with variable-centered approaches that focus on the relationship between individual variables and assume homogeneity among the population of interest with regards to how predictors impact the outcome of interest, person-centered approaches (e.g., latent class analysis [LCA], latent transition analysis [LTA], latent profile analysis [LPA], latent growth mixture model [LGMM]) assume heterogeneity among the population of interest with regards to how outcomes of interest are influenced by predictors. Indeed, these methods examine the data using an inductive lens, analyze participants’ response patterns to certain individual indicator variables that measure specific substance use practices, and reveal unobserved typologies or classes of PSU via statistically transparent and reproducible methods (Lanza & Cooper, 2016; Lanza & Rhoades, 2013).

Given the wide range of substances available and the challenges in measuring PSU in a clinically meaningful manner, it is unclear what types of PSU patterns are prevalent among PWUO. This systematic review aims to classify, characterize, and summarize the latent patterns of PSU among PWUO. Identifying latent PSU classes and associated determinants among PWUO helps facilitate and inform the implementation of evidence-based interventions and preventative measures aimed at addressing the ongoing overdose epidemic.

Methods

Databases and search strategy

Following the Peer Review of Electronic Search Strategies (PRESS) guidelines (McGowan et al., 2016) and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher, Liberati, Tetzlaff, & Altman, 2009), we searched for empirical peer-reviewed studies or gray literature on latent classes of PSU among PWUO from inception, through to June 15, 2020 (See Table S1 for PRISMA checklist). The search concepts, databases searched, and information about the review’s protocol registration are presented in Table 1, and a sample search strategy is available in Table S2.

Table 1. An overview of the search strategy.

Inclusion criteria

Empirical studies of any design (i.e., observational, experimental, cross-sectional, and cohort) were considered for inclusion if they identified latent classes of substance use among PWUO. Opioid use was assessed based on the following criteria: i) participants had regular opioid use; ii) were clinically assessed as having opioid use disorder (OUD) based on The Diagnostic and Statistical Manual of Mental Disorders [DSM] or other validated tools (American Psychiatric Association, 2013); or iii) participants were receiving treatment for OUD. Studies with mixed populations of PWUO and people with other substance use disorders were considered if they provided separate analyses for participants with PWUO or if more than 50% of the participants were PWUO. Studies were included if they had reported categorical or continuous measures of substance use during any time interval (e.g., last year, last six months, last month, current). Studies were only included if they reported latent classes of substance use through various latent class analytical approaches (e.g., LCA, LTA, LPA, LGMM). Eligible studies had to report details about latent classes identified and variables used as outcome indicators (i.e., individual indicators that contributed to identifying latent classes). Studies were excluded if they used latent class analyses for variables other than substance use characteristics (e.g., HIV or overdose risk).

Study selection

Two independent reviewers (MK and AP) completed the title and abstract screening. Studies that met our inclusion criteria or were unclear were retained for full-text screening, which was done by two independent reviewers (MK and AP). Disagreements over the inclusion of studies were resolved through discussion throughout the screening process. Duplicate studies were identified and excluded.

Data extraction and analysis

We developed a data extraction sheet and pilot-tested it by two independent authors (MK and AP). Two reviewers (MK and AP) completed data extraction independently, and discrepancies were resolved through discussion. Data were extracted on study characteristics, participant characteristics, and outcome characteristics. Given the considerable conceptual heterogeneity of covariates (e.g., inclusion of a range of sociodemographic or behavioural variables in the multivariable analyses) and outcomes included in the analyses across the studies (e.g., PSU operationalization using different self-reported or clinical measures as well as the inclusion of various outcome indicators in the latent analyses), no meta-analysis was conducted, and study findings were summarized and presented in a narrative fashion. Primary latent classes and sub-classes of PSU were identified and presented.

Quality assessment

We used a modified version of Newcastle-Ottawa Quality Assessment Scale (Wells et al., 2000) to assess the risk of bias in the included studies independently. The tool uses several components to evaluate selection bias, comparability, and outcome assessment. Given the review’s focus on latent analyses, we also used a modified version of the Guidelines for Reporting on Latent Trajectory Studies (GRoLTS) checklist (Van De Schoot, Sijbrandij, Winter, Depaoli, & Vermunt, 2017) to assess the specific methodological quality of the included studies. The modified GRoLTS checklist (Petersen, Qualter, & Humphrey, 2019) contains 15 yes/no items. Using this checklist is important in critical appraisal of LCA, LPA, or LTA studies and ensuring transparency and interpretability of their results.

Results

Out of the 3372 initial unique records identified, 30 met our inclusion criteria and were included in the systematic review (Fig. 1).

Fig. 1. PRISMA flow diagram of the study screening process.

Study settings and participants

An overview of the included studies is presented in Table 2. Although all studies included a sub-population of PWUO, their definition of opioid use varied greatly and included a range of criteria across the studies (e.g., receiving opioid agonist therapy [OAT] medication, self-reported recent use of illicit and/or non-prescribed opioids, meeting DSM criteria for OUD). Moreover, the inclusion criteria of the individual studies varied considerably and included a wide range of participants, such as people admitted to emergency departments, people receiving OAT, people involved with the justice system, people accessing harm reduction services, social-economically disadvantaged people who inject drugs (PWID), and other PWUO recruited through various large-scale surveys.

Table 2. Overview of included studies in the systematic review of latent polysubstance use among people with opioid use disorder.

While most studies only included substance use outcome indicators, a handful of studies include other indicators: four included socio-demographic factors, such as age, sex, and race, four included mental health-related conditions, such as post-traumatic stress disorder (PTSD), major depressive disorder (MDD), different forms of phobia, and anxiety, one included hospital utilization, and one included physical health, housing status, and lifetime number of overdoses. Moreover, 15 studies included several indicators of a single type of substance (i.e., frequency of use, length of substance use career) in their latent analysis, and 15 studies included an indicator for the route of substance administration (i.e. injection drug use [IDU] vs. non-injection drug use [non-IDU]) in their assessment of classes of substance use patterns. Lastly, two studies focused on specific brands, such as OxyContin and Tylenol, and only two studies included a measure of fentanyl use as a unique indicator in their LCA.

Latent classes of PSU

The latent classes identified in each study are presented in Table 3. The median number of classes was four, ranging from two to 15 classes. In detail, two studies identified two classes, eight studies identified three classes, seven studies identified four classes, five studies identified five classes, three studies identified six classes, four studies identified seven or eight classes, and one study reported 15 classes. The study with 15 classes reported heroin use conditional on membership in various trajectories of other substance types (Eastwood, Strang, & Marsden, 2019). Only four studies grouped their analyses by another unique characteristic of the participants: being on OAT, urbanicity, and race. Although the classes identified across the studies varied considerably, the following latent classes, presented in Table 4, were found in most studies: Infrequent/low PSU; PSU primarily involving heroin use; PSU primarily involving heroin and stimulant use; PSU primarily involving stimulant use; and Frequent PSU.

Table 4. Latent polysubstance use classes among people who use opioids across the studies included in the review.

_{Note: The identified latent polysubstance use classes and sub-classes are among people who use opioids (PWUO). PWUO was defined as those who were using opioids on a regular basis, were assessed to have opioid use disorder, or were receiving treatment for opioid use disorder.}

While most identified classes fit into the overarching latent classes presented above, some were difficult to fit into a particular group due to variations in measurements and indicators used to build classes. For example, including socio-demographic variables and mental health-related conditions as outcome indicators, was reflected in the final class solutions identified in four and three studies, respectively. Moreover, Afshar et al. (2019) identified classes reflecting the intensity of hospital utilization. Furthermore, eight studies labeled at least one class to indicate a considerable level of alcohol use and five labeled at least one class to indicate a considerable level of cannabis use. In addition, 11 studies labeled at least one class that indicated a considerable level of POs or non-opioid prescription medications. Two studies also identified at least one class suggesting significant tobacco use among their participants and six studies identified at least one class that indicated considerable use of prescribed or non-prescribed OAT medications.

Predictors of latent class membership

Comparisons about latent class memberships need to be interpreted with caution due to the heterogeneity of PSU operationalization across different studies. Of the 30 included studies, 18 reported detailed effect size measures for predictors of latent class membership. Most studies compared their PSU class with lower frequency classes, the details of which are presented in Table 3.

Sociodemographic predictors

Most studies compared the classes based on their sociodemographic characteristics; however, most examined comparisons based on age, sex/gender, and race/ethnicity. The findings of the studies regarding the association of age and substance use class membership were relatively consistent and increasing age was often associated with belonging to higher frequency or severity classes; however, a subset of studies reported no significant association between class membership and age.

Among 28 studies that reported on participants’ sex, no study reported any details about participants’ gender, and some confused sex with gender. Overall, the association of sex and latent classes of PSU were inconsistent. For example, compared to males, females were more likely to belong to adverse mental health, opioid, tobacco, cannabis use disorder classes, amphetamine-type stimulant polydrug use class, low-frequency POs use and depressed class, non-opioid prescription drug use class, non-opioids and benzodiazepine classes, polydrug and polyroute users class, and methamphetamine and heroin class. Conversely, compared to females, males were more likely to belong to the following classes: primarily alcohol, buprenorphine, and benzodiazepine use, very high-frequency POs, polysubstance, and elevated psychopathology, cannabis and/or cocaine use, PSU and heroin overdose, heroin and methamphetamine injectors, infrequent POs and heroin, and polyroute stimulant use. Several studies reported no significant association between class membership and sex.

Among 16 studies that reported on the race/ethnicity of the participants, 11 reported detailed associations between this variable and class memberships. Overall, the findings regarding the association of race/ethnicity and PSU classes were inconsistent and studies had used various racial/ethnic classifications in their assessments. For example, compared to non-Whites, White people were more likely to belong to buprenorphine use class, less likely to belong to polydrug use class, and more likely to be classified as high severity users. Conversely, compared to White people, Hispanic and Black people were more likely to belong to heroin IDU class, Black people were more likely to be classified as crack/nasal heroin users, and American Indigenous people were more likely to belong to elevated POs, alcohol-tobacco-cocaine, bipolar, and polysubstance/very high psychopathology classes. Some studies reported no significant association between class membership and race/ethnicity.

PSU characteristics among people receiving OAT

A measure of prescribed OAT medications (e.g., methadone, buprenorphine) was included in four studies, two of which presented stratified analyses based on receipt of OAT. Betts et al., stratified PSU patterns among PWID based on OAT status and reported that among those on OAT, the classes identified reflected co-use of multiple drugs in addition to OAT medication. However, the probability of intensive PSU was higher among those who were not receiving OAT. Gjersing et al., also stratified their analysis based on OAT status and reported those on OAT to be less likely to practice PSU. Moreover, those engaging in PSU among non-OAT receiving PWID were at an increased risk of premature mortality. Daniulaityte et al., characterized the heterogeneous patterns of prescribed and non-prescribed buprenorphine among people with OUD and showed that those receiving formal addiction treatment to be less likely to report intensive non-prescribed buprenorphine. Notably, they highlighted that non-prescribed buprenorphine use among the participants was primarily driven by self-treatment practices to deal with withdrawal symptoms and much less for euphoric purposes. Lastly, Peacock et al., showed that among those with frequent prescribed OAT use, other illicit drugs and non-prescribed opioids were used less and infrequently compared to those not receiving prescribed OAT.

Outcomes associated with latent class membership

Among studies that reported outcomes associated with latent class membership, in comparison with lower frequency groups, belonging to higher frequency or intensity PSU classes were associated with increased odds of various behaviours, such as use of psychiatric services, mental health comorbidities, poor drug-related outcomes, frequent IDU, receptive needle and paraphernalia sharing, thrombosis, engaging in violent behaviours, non-fatal overdose, high-risk sexual behaviours, sexualized substance use, as well as experiences of adversities, such as homelessness, incarceration, unemployment, chronic pain, and HCV sero-positivity. Mortality risk was only assessed in one cohort study of street- and low-threshold service-recruited people who engaged in PSU practices in Norway, where membership in the polysubstance injectors and low-frequency injector classes not receiving OAT, was associated with an increased hazard of mortality in comparison to frequent buprenorphine users (Gjersing & Bretteville-Jensen, 2018).

Quality of the evidence

As presented in Table S3, most studies were of satisfactory quality in terms of risk of bias as evaluated by a modified version of Newcastle-Ottawa Quality Assessment Scale. While most studies had a sufficient sample size required for LCA (Nylund-Gibson & Choi, 2018), had comparable participants, and were statistically sound, most suffered from measurement biases in their outcome and exposure ascertainment. The quality of LCA studies, which was assessed by a modified GRoLTS checklist, are presented in Table S4 and identified several limitations across the included studies.

Discussion

We systematically reviewed 30 studies that applied a latent analytical approach to identifying PSU patterns among PWUO and identified five distinct PSU patterns: Infrequent/low PSU in 16 studies, PSU primarily involving heroin use in 22 studies, PSU primarily involving heroin and stimulant use in 15 studies, PSU primarily involving stimulant use in 13 studies, and frequent PSU in almost all included studies. Membership in higher frequency or intensity PSU classes was associated with several individual- and structural-level adverse mental and physical health outcomes.

Our findings of PSU patterns among PWUO are relatively comparable but not compatible with a previous review of PSU patterns in the general population, which classified PSU into clusters, including no or limited use (alcohol, tobacco, and cannabis), moderate use (“limited range” and amphetamines), and extended use (“moderate range”, illicit prescription medications, and other illicit substances) (Connor et al., 2014). Our findings also diverge from several previous reviews on PSU clusters/classes among adolescents, which mainly identified different subgroups of adolescents using varying frequencies and amounts of tobacco, alcohol, and cannabis (Halladay et al., 2020; Tomczyk et al., 2016). These differences could be attributed to the dissimilar approaches in PSU measurement as well as distinctive behavioural characteristics and substance use patterns of PWUO in comparison with people in the general population or adolescents who may be primarily experimenting with PSU. Future research could benefit from improving the instrumentation of and screening for PSU in clinical settings. Developing pilot-tested and standardized screening tools for PSU that can measure different patterns of PSU, routes of administration (e.g., injection, non-injection), reasons for practicing PSU (e.g., recreational, withdrawal management), and recent experiences (e.g., last week, current) can be beneficial to addiction research and clinical practice. Such screening tools could help increase the understanding of PSU, aid in better identification of people engaged in higher-risk practices, help define meaningful treatment-related outcomes (i.e., client-centered non-abstinence focused), and tailor care and treatment plans accordingly.

Although our results suggested certain distinct classes of PSU among PWUO, PSU seems to be often the norm rather an occasional practice or an exception among a minority group of PWUO. Indeed, our findings are in line with a growing body of evidence that suggests concurrent and/or sequential use of multiple substances, including but not limited to stimulants, opioid/non-opioid prescription medications, and alcohol, is a common practice among PWUO (Cicero, Ellis, & Kasper, 2020; Crummy et al., 2020; Jones, Mogali, & Comer, 2012; Witkiewitz & Vowles, 2018). This established body of evidence points to a range of recreational or therapeutic reasons for PSU among PWUO. First, combining substances that have dissimilar impacts on CNS (e.g., opioids and benzodiazepines or stimulants and opioids, such as goofballs and speedballs) could magnify their euphoric effects (Crummy et al., 2020; Darke, Duflou, Torok, & Prolov, 2013; Jones et al., 2012; Timko et al., 2018; Wang, Chen, Chen, Chou, & Chou, 2014; Witkiewitz & Vowles, 2018). Second, stimulants may be used concurrently or consecutively with heroin to reduce the undesirable experiences of opioid cravings or withdrawal (Harris et al., 2013), or balance out the effects of the heroin high (Al-Tayyib, Koester, Langegger, & Raville, 2017; Ellis, Kasper, & Cicero, 2018; McNeil et al., 2020). Lastly, PSU among PWUO could also be driven by substance use availability in the illicit drug supply. For example, the reduced availability of heroin in 2000 in Australia was associated with a decrease in the number of people who injected heroin and an increase in amphetamine injecting among them (Day, Degenhardt, & Hall, 2006). Regardless of the motivations behind engaging in PSU, our findings are in line with the existing evidence suggesting PSU among PWUO to be associated with several adverse mental and physical health outcomes (e.g., major depressive disorders, PTSD, risky sexual practices) in comparison with those who use few or no substances other than opioids (Connor et al., 2014; Crummy et al., 2020; Jones et al., 2012; Timko et al., 2018; Trenz et al., 2013; Witkiewitz & Vowles, 2018).

Our findings are of particular importance in the context of the ongoing opioid epidemic in the US and Canada, where research studies are often “opioid-centric” and may “miss the forest for the trees” by viewing OUD in a silo and excluding people with several high-frequency uses of multiple substances (Cicero et al., 2020). Policies aimed at tackling the opioid epidemic are also often disproportionately focused on the effectiveness of and access to OUD treatment (e.g., various OAT and antagonist therapies), opioid-overdose prevention (e.g., take-home naloxone kits, fentanyl drug checking services), and preventing access to non-medical POs (e.g., prescription drug monitoring) (Cicero et al., 2020; Finley et al., 2017; Kerr, 2019; McDonald & Strang, 2016; Tupper, McCrae, Garber, Lysyshyn, & Wood, 2018). This exclusive focus on PWUO and a single drug class may be justified by the fact that the drugs of choice among most PWUO are arguably natural (e.g., heroin and its derivatives) or synthetic opioids (e.g., fentanyl and its analogues) (Gryczynski et al., 2019; Kanouse & Compton, 2015; Karamouzian et al., 2020; Ostling et al., 2018). While continued support of these life-saving interventions is essential, our findings suggest that the extent of interventions’ effectiveness for certain subpopulations of PWUO (e.g., those who primarily use stimulants, have concurrent alcohol use disorder, or benzodiazepine use disorder as well as people using other substances while on OAT) might be limited. Recognizing these complexities and considering them in developing care packages for PWUO are essential in addressing the unique needs of these particular subpopulations. These heterogeneities are not just reflected in PWUO’s individual- and structural-level risk factors, but also adverse health outcomes (e.g., fatal and non-fatal OD). These considerable diversities should be reflected in addiction research which often focuses on mono-substance use practices as well as people’s continuum of care in the healthcare system. Implementing policies and practices that are not client-centered, put all PWUO into a single box, and simplify their substance use practices by focusing on their primary drug of choice/use, fail to acknowledge the specific needs and risks of different subgroups among PWUO and may be of limited success.

Predictors of membership in different PSU classes varied significantly across different studies. Among studies that did assess these predictors, membership in higher intensity or frequency of PSU was positively associated with a range of individual- (e.g., sharing needles, frequent IDU, high-risk sexual behaviours), and socio-structural-level exposures (e.g., history of incarceration, homelessness, hospitalization). Study findings regarding the sociodemographic predictors of class membership, however, were relatively inconsistent. While these assessments are informative in characterizing the negative or positive predictors of PSU classes, these findings need to be interpreted with caution for several reasons. First and foremost, PSU operationalization and the proportion of classes identified, varied significantly across the studies. For example, studies used several approaches to measuring patterns of substance use and included an array of different substances from tobacco to fentanyl. Assessments were rarely based on validated metrics and mainly according to arbitrary self-reported measures in the past few days, weeks, months, lifetime, or a combination of several different timelines. Moreover, several studies included multiple indicators of a single type of substance (e.g., frequency, length, and route of use for a certain substance). Additionally, some studies’ identification of various classes of PSU was informed by inclusion of sociodemographic or mental-health-related outcome indicators. These heterogeneities were often reflected in the final class solutions and led to challenges in summarizing the evidence or comparing classes across individual studies. Second, most studies assessed cross-sectional associations between particular behavioural characteristics of the participants with PSU classes, which could raise concerns about temporality bias and establishment of causality. Lastly, several studies did not assess these associations and reported their final class solutions in a merely descriptive fashion and, therefore, provided a limited picture of potential predictors of class membership in their analysis.

Limitations

While our systematic review was methodologically rigorous and comprehensive, it is subject to certain limitations, primarily due to the methodological shortcomings of the included individual studies. First, while all studies included PWUO, the sampling framework, sociodemographic, and behavioural characteristics of the participants were heterogeneous. Second, studies operationalization of PSU patterns varied across the studies and made it challenging to make direct comparisons and estimate pooled effect measures. We were also unable to compare studies that used various outcome indicators in terms of their findings. Moreover, although PSU classes with high frequency of benzodiazepines and POs were not identified as primary classes in the analysis, it is important to note that subgroups of people who frequently use benzodiazepines and POs do exist among PWUO and their specific needs should be recognized in clinical practice and intervention developments. Third, we only focused on latent analytical approaches and did not include studies that used traditional clustering approaches (e.g., k-means, hierarchical cluster analysis) in our review. This decision was informed by latent class analytical approaches’ superior statistical robustness and flexibility to cluster-based models, due to their probabilistic model-based nature (DiStefano & Kamphaus, 2006). Moreover, making methodological comparisons between the studies and summarizing the evidence from two distinct statistical approaches would have been challenging. Fourth, most of the participants in the studies were male, White, and from North America, limiting the findings’ generalizability to other contexts where OUD is a significant public health concern (e.g., West and Southeast Asia). Additionally, given the inconsistencies in classifying race/ethnicities across the studies, we could not provide more detailed assessments about the association of race/ethnicity with PSU classes. Lastly, the findings of these studies should be interpreted with an eye to the rapidly changing illicit opioid supply across the world, in North America in particular. Future studies on measuring PSU among PWUO would benefit from looking at how PSU classes may have evolved in the context of an increasingly toxic drug supply where PWUO may be knowingly or accidentally exposed to synthetic opioids. Despite these methodological limitations, most studies were robust, had a large sample size, and were of reasonable quality. Overall, our systematic review provides an overview of various classes and predictors of PSU patterns among PWUO and helps inform the clinical and public health interventions aimed at addressing the opioid epidemic.

Methodological issues

Our review shows that using latent class and person-centered approaches to identifying distinct PSU patterns among various subgroups of PWUO is useful and informative. However, our quality assessment of the studies identified several areas for improvement for future studies that aim to apply this method to subpopulations of people who use drugs. Most studies used unvalidated or unstandardized measurement approaches for substance use indicators. Another important limitation of PSU analyses across most studies was overlooking tobacco use in their PSU analyses, despite the significant frequency and association of tobacco smoking and opioid use among PWUO (Akerman et al., 2015; John et al., 2019; Rajabi, Dehghani, Shojaei, Farjam, & Motevalian, 2019). Several studies also did not report how missing data were handled in their study. Further clarity and consistency about all possible models’ fit indices and final model selection procedures are essential in creating reproducible, reliable, and comparable results. Moreover, the notable limitations about sex- and gender-based reporting and analytical approaches observed across the studies (e.g., not reporting data on participants’ sex or gender, not conducting sex- or gender-stratified analyses, and confusing sex with gender) further highlight an important knowledge gap (Greaves, 2020), as well as the need for a systematic integration and exploration of sex and gender in future research on latent PSU patterns (Heidari, Babor, De Castro, Tort, & Curno, 2016). Some of the above-mentioned shortcomings have been repeatedly reported in previous reviews aimed at identifying clusters or classes of substance use (Halladay et al., 2020; Petersen et al., 2019; Tomczyk et al., 2016; Van De Schoot et al., 2017) and could be addressed by following the existing reporting guidelines for these types of analyses (Petersen et al., 2019; Van De Schoot et al., 2017).

Conclusions

Our systematic review summarized the evidence that applied person-centered approaches to the classification of PSU among PWUO. While heroin and its derivatives were the most common class, we found that PSU was the norm, not the exception. However, the heterogeneities among PSU classes of PWUO were considerable. Our findings call for further investments and research in developing treatments and interventions that go beyond focusing on the use of opioids among this population and apply a holistic and comprehensive approach to providing care for PWUO. Applying methodologically rigorous and transparent techniques as well as using standardized metrics for evaluating the frequency and severity of substance use patterns are required to allow direct comparisons across the studies’ findings and improve their generalizability.

Introduction

There is no single definition for polysubstance use (PSU); it generally refers to using two or more different substances, or types of substances, at the same time or over a period. Evidence from around the world shows that PSU is common among people who use opioids (PWUO). Additionally, drug overdose deaths often involve several drugs beyond opioids. For example, in British Columbia, Canada, the drugs most often found in illicit drug toxicity deaths over four years included fentanyl (83%), cocaine (50%), amphetamines (34%), and heroin (15%). Similar trends are seen in the United States, where reported cocaine overdose deaths involving any opioids increased from 29% to 63% between 2000 and 2015.

Measuring PSU is difficult, both statistically and in terms of research methods. Many studies have looked at PSU by comparing individual drug use, which often leads to small data groups or uneven distributions. This can make analyses limited or inaccurate. Even with different definitions of PSU, studies consistently show it is linked to a higher risk of many poor health outcomes.

Understanding PSU patterns is important for overdose prevention and harm reduction. Because of this, new methods have been developed to identify specific groups of people with distinct PSU patterns. These are called person-centered methods. Variable-centered approaches focus on how individual factors relate to outcomes, assuming everyone in the group is similar in how these factors affect them. Person-centered approaches, however, assume that people within the group are different in how factors influence outcomes. These methods use data to discover hidden types or classes of PSU. They do this by analyzing how participants respond to questions about their substance use practices, using clear and repeatable statistical techniques.

With many substances available and the difficulty in measuring PSU in a way that is useful for clinics, it is not clear what types of PSU patterns are common among PWUO. This systematic review aims to identify, describe, and summarize the hidden patterns of PSU among PWUO. Knowing these hidden PSU classes and what factors are linked to them helps guide effective interventions and prevention efforts to address the ongoing overdose crisis.

Methods

Databases and search strategy

The search followed guidelines for systematic reviews, including the Peer Review of Electronic Search Strategies (PRESS) and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. Studies were searched from their beginning until June 15, 2020. The search concepts, databases, and review protocol registration information are detailed in related tables.

Inclusion criteria

Studies of any design (observational, experimental, cross-sectional, or cohort) were included if they identified hidden classes of substance use among PWUO. Opioid use was defined by: 1) regular opioid use, 2) a clinical diagnosis of opioid use disorder (OUD) using standard tools like the Diagnostic and Statistical Manual of Mental Disorders (DSM), or 3) receiving treatment for OUD. Studies with mixed groups of PWUO and people with other substance use disorders were included if they analyzed PWUO separately or if more than half of participants were PWUO. Studies were included if they reported substance use as categories or continuous measures over any time period (e.g., last year, last six months, last month, or current). Only studies that reported hidden classes of substance use using methods like latent class analysis (LCA), latent transition analysis (LTA), latent profile analysis (LPA), or latent growth mixture models (LGMM) were included. Eligible studies needed to describe the hidden classes found and the variables used to identify them. Studies were excluded if they used latent class analysis for factors other than substance use (e.g., HIV or overdose risk).

Study selection

Two independent reviewers screened titles and abstracts. Studies that met inclusion criteria or were unclear were kept for full-text screening, also done by two independent reviewers. Disagreements about study inclusion were resolved through discussion. Duplicate studies were identified and removed.

Data extraction and analysis

A data extraction sheet was developed and tested by two independent authors. Two reviewers completed data extraction independently, and discrepancies were resolved through discussion. Data was extracted on study characteristics, participant characteristics, and outcome characteristics. Due to the wide variety of factors and outcomes across studies, a meta-analysis was not possible. Study findings were summarized descriptively. Primary hidden classes and sub-classes of PSU were identified and presented.

Quality assessment

A modified Newcastle-Ottawa Quality Assessment Scale was used to independently assess the risk of bias in the included studies. This tool uses several components to evaluate selection bias, comparability, and outcome assessment. A modified Guidelines for Reporting on Latent Trajectory Studies (GRoLTS) checklist was also used to assess the specific methodological quality of the studies. This checklist is important for evaluating LCA, LPA, or LTA studies, ensuring their results are clear and easy to understand.

Results

From 3372 unique records, 30 met the inclusion criteria and were included in the review.

Study settings and participants

While all studies focused on PWUO, their definitions of opioid use varied widely. The inclusion criteria for individual studies also varied greatly, including a wide range of participants. These included people admitted to emergency departments, people receiving opioid agonist therapy (OAT), people involved with the justice system, people accessing harm reduction services, socio-economically disadvantaged people who inject drugs (PWID), and other PWUO recruited through various large-scale surveys.

Most studies looked at substance use outcomes. However, some also included other factors: four included demographic factors like age, sex, and race; four included mental health conditions like post-traumatic stress disorder (PTSD) and major depressive disorder (MDD); one included hospital use; and one included physical health, housing, and lifetime overdoses. Additionally, 15 studies included multiple measures for a single substance (e.g., frequency of use, length of use), and 15 studies included an indicator for the route of administration (e.g., injection vs. non-injection drug use). Only two studies measured specific brands like OxyContin and Tylenol, and only two included fentanyl use as a distinct indicator in their LCA.

Latent classes of PSU

The studies identified a median of four hidden classes, with a range from two to 15 classes. Specifically, two studies found two classes, eight found three, seven found four, five found five, three found six, four found seven or eight, and one study reported 15 classes. The study with 15 classes looked at heroin use based on patterns of other substance use. Only four studies organized their analyses by other participant traits, such as being on OAT, living in urban areas, or race. Despite significant variations, most studies identified these common hidden classes: Infrequent/low PSU; PSU mainly involving heroin; PSU mainly involving heroin and stimulants; PSU mainly involving stimulants; and Frequent PSU.

While most classes fit these broad categories, some were hard to classify due to differences in measurements and indicators. For example, the inclusion of demographic variables and mental health conditions influenced the final class solutions in four and three studies, respectively. One study identified classes reflecting the intensity of hospital use. Furthermore, eight studies labeled at least one class indicating a considerable level of alcohol use, and five labeled at least one class indicating a considerable level of cannabis use. In addition, 11 studies labeled at least one class indicating a considerable level of prescription opioids (POs) or non-opioid prescription medications. Two studies also identified at least one class suggesting significant tobacco use, and six studies identified at least one class indicating considerable use of prescribed or non-prescribed OAT medications.

Predictors of latent class membership

Comparisons of hidden class memberships should be made carefully because PSU was defined differently across studies. Eighteen of the 30 studies provided detailed measures for factors predicting hidden class membership. Most studies compared their PSU classes with lower frequency classes.

Sociodemographic predictors

Most studies compared classes based on age, sex/gender, and race/ethnicity. Findings on age were consistent: increasing age was often linked to higher frequency or severity classes. However, some studies found no significant link between class membership and age.

Among 28 studies reporting on participant sex, none detailed gender, and some studies confused sex with gender. Overall, the link between sex and hidden classes of PSU was inconsistent. For example, females were more likely than males to be in classes associated with adverse mental health, opioid, tobacco, cannabis use disorder, amphetamine-type stimulant polydrug use, low-frequency POs use and depression, non-opioid prescription drug use, non-opioids and benzodiazepine classes, polydrug and polyroute users, and methamphetamine and heroin. Conversely, males were more likely than females to be in classes associated with primarily alcohol, buprenorphine, and benzodiazepine use, very high-frequency POs, polysubstance, and elevated psychopathology, cannabis and/or cocaine use, PSU and heroin overdose, heroin and methamphetamine injectors, infrequent POs and heroin, and polyroute stimulant use. Several studies found no significant link between class membership and sex.

Findings on race/ethnicity and PSU classes were inconsistent, with studies using various classifications. For example, White individuals were more likely than non-White individuals to be in the buprenorphine use class, less likely to be in the polydrug use class, and more likely to be classified as high severity users. Conversely, Hispanic and Black individuals were more likely than White individuals to be in the heroin injection drug use class, Black individuals were more likely to be classified as crack/nasal heroin users, and American Indigenous individuals were more likely to be in elevated POs, alcohol-tobacco-cocaine, bipolar, and polysubstance/very high psychopathology classes. Some studies reported no significant link between class membership and race/ethnicity.

PSU characteristics among people receiving OAT

Four studies included prescribed OAT medications (e.g., methadone, buprenorphine), and two of these analyzed data based on whether participants received OAT. One study showed that among people who inject drugs on OAT, classes reflected co-use of multiple drugs. However, intensive PSU was more likely among those not receiving OAT. Another study found that people on OAT were less likely to engage in PSU. Among those who inject drugs and were not on OAT, engaging in PSU was linked to a higher risk of early death. One study noted that non-prescribed buprenorphine use was mainly for self-treatment of withdrawal symptoms, not for euphoric effects. Another study indicated that frequent prescribed OAT use was associated with less frequent use of other illicit drugs and non-prescribed opioids, compared to not receiving prescribed OAT.

Outcomes associated with latent class membership

Studies reporting outcomes linked to hidden class membership found that higher frequency or intensity PSU classes were associated with increased likelihood of various issues. These included using psychiatric services, having mental health problems, poor drug-related outcomes, frequent injection drug use, sharing needles, thrombosis, violent behaviors, non-fatal overdose, high-risk sexual behaviors, and sexualized substance use. Also linked were adversities like homelessness, incarceration, unemployment, chronic pain, and HCV positivity.

Only one study, from Norway, assessed mortality risk. It found that belonging to polysubstance injector or low-frequency injector classes not receiving OAT was linked to a higher risk of death, compared to frequent buprenorphine users.

Quality of the evidence

Most studies had satisfactory quality regarding bias risk, as evaluated by a modified Newcastle-Ottawa Quality Assessment Scale. While most studies had sufficient sample sizes for LCA, comparable participants, and were statistically sound, many had measurement biases in how outcomes and exposures were determined. The quality of LCA studies, assessed by a modified GRoLTS checklist, showed several limitations across the included studies.

Discussion

This systematic review examined 30 studies using hidden class analysis to identify PSU patterns among PWUO. Five distinct PSU patterns were found: Infrequent/low PSU; PSU mainly involving heroin; PSU mainly involving heroin and stimulants; PSU mainly involving stimulants; and Frequent PSU. Belonging to higher frequency or intensity PSU classes was linked to various negative mental and physical health outcomes, affecting individuals and their broader circumstances.

The PSU patterns found among PWUO are somewhat similar to, but not entirely consistent with, a previous review of PSU in the general population. That review classified PSU into clusters like no/limited use, moderate use, and extended use. The findings also differ from reviews on PSU classes among adolescents, which largely identified subgroups based on varying frequencies of tobacco, alcohol, and cannabis use. These differences may be due to different ways of measuring PSU, and the distinct behaviors and substance use patterns of PWUO compared to the general population or adolescents who might be experimenting. Future research could improve how PSU is measured and screened for in clinical settings. Creating standardized and tested screening tools for PSU would be helpful for addiction research and clinical practice. These tools could measure different patterns, routes of administration (e.g., injection), reasons for use (e.g., recreation, withdrawal management), and recent experiences. Such tools could improve understanding of PSU, help identify people at higher risk, define relevant treatment outcomes (e.g., non-abstinence focused), and customize care plans.

While distinct PSU classes were identified among PWUO, PSU appears to be common, rather than an infrequent or rare practice. These findings align with growing evidence that using multiple substances concurrently or sequentially—including stimulants, prescription medications (opioid and non-opioid), and alcohol—is common among PWUO. Reasons for combining substances include magnifying euphoric effects (e.g., opioids and benzodiazepines, or stimulants and opioids), reducing opioid cravings or withdrawal, or balancing the effects of a heroin high. PSU among PWUO can also be influenced by the availability of substances in the illicit drug supply. Regardless of reasons for PSU, findings consistent with existing evidence show PSU among PWUO is linked to several negative mental and physical health outcomes (e.g., major depressive disorders, PTSD, risky sexual practices), especially when compared to those who use few or no other substances besides opioids.

These findings are especially relevant to the ongoing opioid crisis in the US and Canada. Research often focuses solely on opioids, potentially overlooking the broader picture by studying OUD in isolation and excluding people who frequently use multiple substances. Policies addressing the opioid epidemic often focus heavily on OUD treatment effectiveness and access (e.g., various OAT and antagonist therapies), opioid overdose prevention (e.g., take-home naloxone kits, fentanyl drug checking services), and preventing access to non-medical prescription opioids. This singular focus might be justified because the primary drugs of choice for most PWUO are often natural or synthetic opioids. While these life-saving interventions are crucial, the findings suggest their effectiveness may be limited for certain PWUO subgroups (e.g., those primarily using stimulants, or with concurrent alcohol or benzodiazepine use disorder, or using other substances while on OAT). Acknowledging these complexities and incorporating them into care plans for PWUO is essential to meet the unique needs of these specific subgroups. These differences appear not only in individual and societal risk factors for PWUO, but also in negative health outcomes (e.g., fatal and non-fatal overdose). Addiction research, which often focuses on single substance use and people's healthcare journey, should reflect these diverse patterns. Policies and practices that are not client-centered, treat all PWUO similarly, and simplify their substance use by focusing only on their primary drug, fail to address the specific needs and risks of different subgroups of PWUO and may have limited success.

Factors predicting membership in different PSU classes varied significantly across studies. Among studies that assessed these predictors, belonging to higher intensity or frequency PSU classes was linked to various individual-level factors (e.g., sharing needles, frequent injection drug use, high-risk sexual behaviors) and socio-structural exposures (e.g., history of incarceration, homelessness, hospitalization). However, findings on sociodemographic predictors of class membership were inconsistent. While these assessments help describe the factors predicting PSU classes, the findings should be interpreted with caution for several reasons. First, PSU definition and the proportion of classes found varied significantly. Studies used various ways to measure substance use patterns, including a range of substances from tobacco to fentanyl. Assessments were rarely based on validated measures and mainly relied on self-reported data from different timeframes (e.g., past few days, weeks, months, or lifetime). Some studies also included multiple indicators for a single substance (e.g., frequency, duration, and route of use). Additionally, some studies identified PSU classes based on the inclusion of demographic or mental health outcome indicators, which often influenced the final class solutions. Second, most studies assessed cross-sectional links between participant characteristics and PSU classes. This raises concerns about establishing cause-and-effect relationships due to potential temporality bias. Lastly, several studies did not assess these links and only described their final class solutions, providing limited information on potential predictors of class membership.

Limitations

While this systematic review was thorough, it has limitations, mainly due to methodological weaknesses in the included studies. First, while all studies included PWUO, the sampling methods, demographic, and behavioral characteristics of participants were diverse. Second, how studies defined PSU patterns varied, making direct comparisons and pooled effect measures challenging. Comparing findings from studies using different outcome indicators was not possible. Furthermore, although high-frequency benzodiazepine and prescription opioid PSU classes were not primary findings, it is important to recognize that subgroups of PWUO frequently use these substances, and their specific needs require attention in clinical practice and intervention development. Third, only latent analytical approaches were included, excluding traditional clustering methods (e.g., k-means, hierarchical cluster analysis). This choice was based on the statistical robustness and flexibility of latent class analytical approaches compared to cluster-based models, due to their probabilistic nature. Comparing methods and summarizing evidence from two distinct statistical approaches would have also been challenging. Fourth, most participants were male, White, and from North America, which limits the generalizability of findings to other regions where OUD is a major public health issue (e.g., West and Southeast Asia). Also, inconsistencies in race/ethnicity classification across studies prevented more detailed assessments of its association with PSU classes. Finally, study findings should be considered in light of the rapidly changing illicit opioid supply worldwide, especially in North America. Future studies on measuring PSU among PWUO should examine how PSU classes may have changed due to an increasingly toxic drug supply, where PWUO might be exposed to synthetic opioids, either intentionally or accidentally. Despite these methodological limitations, most studies were robust, had large sample sizes, and were of reasonable quality. Overall, this systematic review provides an overview of PSU patterns and their predictors among PWUO, informing clinical and public health interventions for the opioid epidemic.

Methodological issues

The review shows that using latent class and person-centered approaches is useful for identifying distinct PSU patterns among various subgroups of PWUO. However, the quality assessment identified several areas for improvement in future studies applying this method to drug-using populations. Most studies used unvalidated or unstandardized ways to measure substance use. A notable limitation was that most PSU analyses overlooked tobacco use, despite its frequent link with opioid use among PWUO. Several studies also did not report how they handled missing data. More clarity and consistency on model fit indices and final model selection procedures are needed for reproducible, reliable, and comparable results. Moreover, limitations in sex- and gender-based reporting and analysis (e.g., not reporting sex/gender, not stratifying analyses, confusing sex with gender) highlight a knowledge gap and the need to systematically include sex and gender in future research on hidden PSU patterns. Some of these shortcomings have been reported in previous reviews of substance use clusters or classes and could be addressed by following existing reporting guidelines for these types of analyses.

Conclusions

This systematic review summarized evidence using person-centered approaches to classify PSU among PWUO. While heroin and its derivatives formed the most common class, PSU was found to be the norm, not the exception. However, significant variations existed among the PSU classes of PWUO. The findings suggest a need for further investment and research into treatments and interventions that extend beyond a sole focus on opioid use, instead applying a holistic approach to care for PWUO. Using methodologically rigorous and transparent techniques, alongside standardized metrics for evaluating the frequency and severity of substance use patterns, is necessary for direct comparisons across study findings and to improve their generalizability.

Introduction

Defining polysubstance use (PSU) is often challenging, as there is little agreement on its exact meaning. The term broadly describes using two or more different substances or types of substances. This can involve using them at the same time or separately over a specific period. A growing amount of evidence from around the world shows that PSU is especially common among individuals who use opioids. Also, many drug overdose deaths involve several drugs in addition to opioids.

For instance, in the Canadian province of British Columbia, common drugs found in illicit drug toxicity deaths over the past four years included fentanyl, cocaine, amphetamines, and heroin. Similar patterns are seen in the United States, where reports indicate that cocaine overdose deaths involving any opioids significantly increased between 2000 and 2015.

Measuring PSU presents statistical and methodological difficulties. Most studies have assessed PSU by examining combinations of single drugs, which can lead to small data sets or uneven distributions. This can limit or bias how analyses are performed or interpreted. Despite disagreements about what constitutes PSU, studies consistently link it to a higher risk of many negative health outcomes.

Understanding the distribution and related factors of PSU is important for overdose prevention and harm reduction programs. Because of this, new methods have been developed to better identify and uncover groups of people with distinct PSU patterns. These "person-centered" methods differ from traditional "variable-centered" approaches. Variable-centered methods focus on the relationships between individual variables and assume that a population responds similarly to predictors. In contrast, person-centered methods assume that a population is diverse in how outcomes are affected by predictors. These methods look at how people respond to different substance use indicators to reveal unobserved types or classes of PSU through clear and repeatable statistical techniques.

Given the wide variety of available substances and the challenges in measuring PSU in a way that is meaningful for clinical practice, it is unclear what types of PSU patterns are common among individuals who use opioids. This systematic review aims to categorize, describe, and summarize these hidden patterns of PSU. Identifying these patterns and their related factors can help develop and implement effective interventions and preventive measures to address the ongoing overdose epidemic.

Methods

Databases and search strategy

To conduct this review, established guidelines for systematic reviews were followed. Empirical peer-reviewed studies or reports on hidden patterns of PSU among individuals who use opioids were searched from the beginning of available records up to June 15, 2020. Information about the search concepts, databases used, and registration of the review's plan are provided in a table.

Inclusion criteria

Studies were included if they were empirical studies of any design (e.g., observational, experimental, cross-sectional, cohort) and identified hidden patterns of substance use among individuals who use opioids. Opioid use was defined by several criteria, such as regular opioid use, a clinical diagnosis of opioid use disorder based on standard guidelines, or receiving treatment for opioid use disorder. Studies with mixed groups of participants were considered if they provided separate analyses for those who use opioids or if more than 50% of participants used opioids. Studies were included if they reported categorical or continuous measures of substance use over any time period (e.g., last year, last six months, last month, current). Only studies that reported hidden patterns of substance use through various latent class analytical approaches were included. Eligible studies also needed to provide details about the identified hidden patterns and the specific variables used to define them. Studies were excluded if they used latent class analyses for variables unrelated to substance use characteristics, such as HIV or overdose risk.

Study selection

Two independent reviewers screened titles and abstracts. Studies that met the inclusion criteria or were unclear were selected for full-text review, also conducted by two independent reviewers. Any disagreements about including studies were resolved through discussion during the screening process. Duplicate studies were identified and removed.

Data extraction and analysis

A data extraction sheet was developed and tested by two independent authors. Two reviewers completed data extraction independently, resolving any differences through discussion. Data were collected on study characteristics, participant characteristics, and outcome characteristics. Due to the significant conceptual differences in the variables used (e.g., various demographic or behavioral factors in analyses) and the different ways PSU was measured and outcomes were defined across studies, a meta-analysis was not performed. Instead, study findings were summarized and presented narratively. The main hidden patterns and sub-patterns of PSU were identified and presented.

Quality assessment

A modified version of a standard quality assessment tool was used to independently evaluate the risk of bias in the included studies. This tool assesses selection bias, comparability, and outcome assessment. Because the review focused on latent analyses, a modified checklist for reporting on latent trajectory studies was also used to assess the specific methodological quality of the included studies. This checklist helps in critically evaluating studies that use latent class analysis and ensures the transparency and interpretability of their results.

Results

Of the initially identified 3372 unique records, 30 met the inclusion criteria and were included in this systematic review.

Study settings and participants

An overview of the included studies is presented in a table. While all studies included a subgroup of individuals who use opioids, the definition of opioid use varied significantly across studies. This included a range of criteria such as receiving opioid agonist therapy, self-reporting recent illicit or non-prescribed opioid use, or meeting diagnostic criteria for opioid use disorder. Furthermore, the criteria for including participants in individual studies varied considerably, encompassing a wide range of individuals such as those admitted to emergency departments, those receiving opioid agonist therapy, individuals involved with the justice system, those accessing harm reduction services, socioeconomically disadvantaged people who inject drugs, and other individuals who use opioids recruited through large-scale surveys.

Most studies only included substance use outcome indicators. However, a few studies included other indicators: four included demographic factors like age, sex, and race; four included mental health conditions such as post-traumatic stress disorder, major depressive disorder, different phobias, and anxiety; one included hospital utilization; and one included physical health, housing status, and lifetime number of overdoses. Additionally, 15 studies included several indicators of a single type of substance (e.g., frequency of use, duration of substance use) in their latent analysis, and 15 studies included an indicator for the route of substance administration (e.g., injection drug use versus non-injection drug use) in their assessment of substance use patterns. Finally, two studies focused on specific brands, such as OxyContin and Tylenol, and only two studies included a measure of fentanyl use as a unique indicator in their latent class analysis.

Latent classes of PSU

The hidden patterns identified in each study are presented in a table. The typical number of classes found was four, with a range from two to 15 classes. Specifically, two studies identified two classes, eight studies identified three classes, seven studies identified four classes, five studies identified five classes, three studies identified six classes, four studies identified seven or eight classes, and one study reported 15 classes. The study with 15 classes reported heroin use that was conditional on belonging to various patterns of other substance types. Only four studies grouped their analyses by another specific characteristic of the participants: being on opioid agonist therapy, urbanicity, and race. Although the classes identified across studies varied considerably, the following hidden patterns were found in most studies: Infrequent/low PSU; PSU primarily involving heroin use; PSU primarily involving heroin and stimulant use; PSU primarily involving stimulant use; and Frequent PSU.

The identified latent polysubstance use classes and sub-classes were found among individuals who use opioids (PWUO). PWUO was defined as those regularly using opioids, assessed as having opioid use disorder, or receiving treatment for opioid use disorder.

While most identified classes fit into the broad hidden patterns described above, some were difficult to categorize due to variations in measurements and indicators used to build the classes. For example, the inclusion of sociodemographic variables and mental health conditions as outcome indicators was reflected in the final class solutions in four and three studies, respectively. Furthermore, one study identified classes reflecting the intensity of hospital utilization. Additionally, eight studies identified at least one class indicating significant alcohol use, and five indicated significant cannabis use. In addition, 11 studies identified at least one class indicating a considerable level of prescription opioids or non-opioid prescription medications. Two studies also identified at least one class suggesting significant tobacco use among their participants, and six studies identified at least one class indicating considerable use of prescribed or non-prescribed opioid agonist therapy medications.

Predictors of latent class membership

Comparisons regarding membership in hidden substance use patterns should be interpreted carefully due to the varied ways PSU was defined and measured across different studies. Of the 30 included studies, 18 provided detailed information on the strength of association for factors predicting membership in a particular hidden pattern. Most studies compared their PSU class with classes that had lower frequency of substance use.

Sociodemographic predictors

Most studies compared the classes based on their demographic characteristics, with age, sex/gender, and race/ethnicity being the most common factors examined. The findings regarding the association of age and substance use class membership were relatively consistent: increasing age was often linked to belonging to classes with higher frequency or severity of substance use. However, a subset of studies reported no significant association between class membership and age.

Among 28 studies that reported on participants’ sex, no study provided details about participants’ gender, and some studies confused sex with gender. Overall, the association between sex and hidden patterns of PSU was inconsistent. For example, females, compared to males, were more likely to belong to classes associated with adverse mental health, opioid, tobacco, or cannabis use disorders, amphetamine-type stimulant polydrug use, low-frequency prescription opioid use and depression, non-opioid prescription drug use, non-opioids and benzodiazepine use, polydrug and polyroute users, and methamphetamine and heroin use. Conversely, compared to females, males were more likely to belong to classes primarily involving alcohol, buprenorphine, and benzodiazepine use, very high-frequency prescription opioid use, polysubstance use with elevated mental health issues, cannabis and/or cocaine use, PSU and heroin overdose, heroin and methamphetamine injectors, infrequent prescription opioid and heroin use, and polyroute stimulant use. Several studies reported no significant association between class membership and sex.

Among 16 studies that reported on the race/ethnicity of the participants, 11 provided detailed associations between this variable and class memberships. Overall, the findings regarding the association of race/ethnicity and PSU classes were inconsistent, and studies used various racial/ethnic classifications in their assessments. For example, White individuals, compared to non-Whites, were more likely to belong to a buprenorphine use class, less likely to belong to a polydrug use class, and more likely to be classified as high severity users. Conversely, Hispanic and Black individuals, compared to White individuals, were more likely to belong to a heroin injection drug use class. Black individuals were more likely to be classified as crack/nasal heroin users, and American Indigenous people were more likely to belong to elevated prescription opioid use, alcohol-tobacco-cocaine use, bipolar disorder, and polysubstance/very high mental health issues classes. Some studies reported no significant association between class membership and race/ethnicity.

PSU characteristics among people receiving OAT

A measure of prescribed opioid agonist therapy (OAT) medications (e.g., methadone, buprenorphine) was included in four studies, two of which presented analyses based on whether participants received OAT. One study found that among individuals who inject drugs, those on OAT had classes that reflected co-use of multiple drugs in addition to OAT medication. However, the likelihood of intensive PSU was higher among those not receiving OAT. Another study also analyzed data based on OAT status and reported that those on OAT were less likely to engage in PSU. Furthermore, among individuals who inject drugs and were not receiving OAT, engaging in PSU was associated with an increased risk of early death. Additionally, one study characterized the varied patterns of prescribed and non-prescribed buprenorphine among individuals with opioid use disorder and showed that those receiving formal addiction treatment were less likely to report intensive non-prescribed buprenorphine use. Notably, that study highlighted that non-prescribed buprenorphine use among participants was mainly driven by self-treatment to manage withdrawal symptoms, and much less for euphoric purposes. Lastly, one study showed that among those with frequent prescribed OAT use, other illicit drugs and non-prescribed opioids were used less and infrequently compared to those not receiving prescribed OAT.

Outcomes associated with latent class membership

Among studies that reported outcomes linked to hidden class membership, belonging to higher frequency or intensity PSU classes was associated with increased likelihood of various behaviors. These included using psychiatric services, having mental health conditions, experiencing poor drug-related outcomes, frequent injection drug use, sharing needles and drug paraphernalia, developing blood clots, engaging in violent behaviors, non-fatal overdose, high-risk sexual behaviors, and substance use during sexual activity. It was also associated with experiencing adversities such as homelessness, incarceration, unemployment, chronic pain, and hepatitis C seropositivity. The risk of death was only assessed in one long-term study of individuals who engaged in PSU practices in Norway. In that study, membership in the polysubstance injectors and low-frequency injector classes not receiving OAT was associated with an increased risk of death compared to frequent buprenorphine users.

Quality of the evidence

As presented in a table, most studies were of satisfactory quality in terms of bias risk, as evaluated by a modified assessment scale. While most studies had a sufficient sample size for latent class analysis, had comparable participants, and were statistically sound, most suffered from measurement biases in how outcomes and exposures were determined. The quality of latent class analysis studies, assessed by a modified checklist, revealed several limitations across the included studies.

Discussion

This systematic review examined 30 studies that used a latent analytical approach to identify PSU patterns among individuals who use opioids. Five distinct PSU patterns were identified: Infrequent/low PSU (in 16 studies), PSU primarily involving heroin use (in 22 studies), PSU primarily involving heroin and stimulant use (in 15 studies), PSU primarily involving stimulant use (in 13 studies), and frequent PSU (in almost all included studies). Membership in higher frequency or intensity PSU classes was linked to several adverse mental and physical health outcomes at both the individual and systemic levels.

The findings regarding PSU patterns among individuals who use opioids are somewhat comparable, but not entirely consistent, with a previous review of PSU patterns in the general population. That review classified PSU into clusters including no or limited use (alcohol, tobacco, and cannabis), moderate use ("limited range" and amphetamines), and extended use ("moderate range," illicit prescription medications, and other illicit substances). These findings also differ from several earlier reviews on PSU clusters/classes among adolescents, which mainly identified different subgroups of adolescents using varying frequencies and amounts of tobacco, alcohol, and cannabis. These differences could be due to varying approaches in measuring PSU, as well as distinct behavioral characteristics and substance use patterns of individuals who use opioids compared to people in the general population or adolescents who may primarily be experimenting with PSU. Future research could benefit from improving the tools and screening methods for PSU in clinical settings. Developing and testing standardized screening tools for PSU that can measure different patterns of use, routes of administration (e.g., injection, non-injection), reasons for practicing PSU (e.g., recreational, withdrawal management), and recent experiences (e.g., last week, current) could be beneficial for addiction research and clinical practice. Such screening tools could enhance understanding of PSU, help better identify individuals engaging in higher-risk practices, aid in defining meaningful treatment outcomes (e.g., client-centered, not solely abstinence-focused), and allow for tailored care and treatment plans.

Although the results suggested certain distinct classes of PSU among individuals who use opioids, PSU appears to be the norm rather than an occasional practice or an exception among a minority of this group. Indeed, these findings align with a growing body of evidence indicating that concurrent and/or sequential use of multiple substances, including but not limited to stimulants, opioid/non-opioid prescription medications, and alcohol, is a common practice among individuals who use opioids. This established evidence points to a range of recreational or therapeutic reasons for PSU among these individuals. First, combining substances that have different effects on the central nervous system (e.g., opioids and benzodiazepines or stimulants and opioids, often referred to as "goofballs" and "speedballs") can magnify their euphoric effects. Second, stimulants may be used concurrently or consecutively with heroin to reduce the undesirable experiences of opioid cravings or withdrawal, or to balance out the effects of the heroin high. Lastly, PSU among individuals who use opioids could also be influenced by the availability of substances in the illicit drug supply. For example, a decrease in heroin availability in Australia in 2000 was linked to a reduction in heroin injection and an increase in amphetamine injection among individuals who inject drugs. Regardless of the motivations behind engaging in PSU, the findings align with existing evidence suggesting that PSU among individuals who use opioids is associated with several adverse mental and physical health outcomes (e.g., major depressive disorders, post-traumatic stress disorder, risky sexual practices) compared to those who use few or no substances other than opioids.

These findings are particularly important in the context of the ongoing opioid epidemic in the United States and Canada. Research studies are often "opioid-centric," potentially overlooking the broader picture by viewing opioid use disorder in isolation and excluding individuals with frequent use of multiple substances. Policies aimed at tackling the opioid epidemic often disproportionately focus on the effectiveness of and access to opioid use disorder treatment (e.g., various opioid agonist therapy and antagonist therapies), opioid-overdose prevention (e.g., take-home naloxone kits, fentanyl drug checking services), and preventing access to non-medical prescription opioids (e.g., prescription drug monitoring). This exclusive focus on individuals who use opioids and a single drug class may be justified by the fact that the primary drugs of choice among most individuals who use opioids are arguably natural (e.g., heroin and its derivatives) or synthetic opioids (e.g., fentanyl and its analogues). While continued support of these life-saving interventions is essential, these findings suggest that the effectiveness of interventions for certain subgroups of individuals who use opioids (e.g., those who primarily use stimulants, have concurrent alcohol use disorder, or benzodiazepine use disorder, as well as individuals using other substances while on opioid agonist therapy) might be limited. Recognizing these complexities and considering them when developing care packages for individuals who use opioids are essential in addressing the unique needs of these particular subgroups. These differences are reflected not just in individuals' and systemic risk factors but also in adverse health outcomes (e.g., fatal and non-fatal overdose). These considerable diversities should be reflected in addiction research, which often focuses on single-substance use practices, as well as in the continuum of care within the healthcare system. Implementing policies and practices that are not client-centered, group all individuals who use opioids into a single category, and simplify their substance use practices by focusing on their primary drug of choice/use, fail to acknowledge the specific needs and risks of different subgroups among individuals who use opioids and may have limited success.

Predictors of membership in different PSU classes varied significantly across different studies. Among studies that did assess these predictors, membership in higher intensity or frequency of PSU was positively associated with a range of individual-level factors (e.g., sharing needles, frequent injection drug use, high-risk sexual behaviors) and socio-structural-level exposures (e.g., history of incarceration, homelessness, hospitalization). Study findings regarding the demographic predictors of class membership, however, were relatively inconsistent. While these assessments are informative in characterizing the negative or positive predictors of PSU classes, these findings need to be interpreted with caution for several reasons. First and foremost, the way PSU was defined and the proportion of classes identified varied significantly across the studies. For example, studies used several approaches to measuring patterns of substance use and included a variety of different substances from tobacco to fentanyl. Assessments were rarely based on validated metrics and mainly relied on arbitrary self-reported measures over different timelines (e.g., past few days, weeks, months, lifetime, or a combination). Moreover, several studies included multiple indicators of a single type of substance (e.g., frequency, duration, and route of use for a certain substance). Additionally, some studies' identification of various classes of PSU was influenced by the inclusion of demographic or mental-health-related outcome indicators. These differences were often reflected in the final class solutions and made it challenging to summarize the evidence or compare classes across individual studies. Second, most studies assessed cross-sectional associations between specific behavioral characteristics of the participants and PSU classes, which could raise concerns about temporal bias and the ability to establish cause and effect. Lastly, several studies did not assess these associations and reported their final class solutions in a purely descriptive manner, providing a limited picture of potential predictors of class membership in their analysis.

Limitations

While this systematic review was thorough and methodologically sound, it has certain limitations, primarily due to the methodological weaknesses of the individual studies included. The studies had varied participant groups in terms of sampling, demographics, and behaviors. The ways studies defined and measured PSU patterns also differed, making direct comparisons difficult and preventing a combined estimate of effects. For instance, while high-frequency benzodiazepine and prescription opioid use classes were not identified as primary categories in the overall analysis, it is important to note that subgroups of people who frequently use these substances do exist among individuals who use opioids, and their specific needs should be recognized in clinical practice and intervention development. The review focused exclusively on latent analytical approaches, excluding studies that used traditional clustering methods, a decision based on the superior statistical robustness and flexibility of latent class analytical approaches. Furthermore, most participants in the included studies were male, White, and from North America, which limits the generalizability of the findings to other regions where opioid use disorder is a significant public health concern. Inconsistencies in classifying race/ethnicities across studies also prevented more detailed assessments of these associations with PSU classes. Finally, the findings should be interpreted considering the rapidly changing illicit opioid supply worldwide, particularly in North America. Future studies on measuring PSU among individuals who use opioids would benefit from examining how PSU patterns may have evolved in the context of an increasingly toxic drug supply, where individuals who use opioids may be knowingly or accidentally exposed to synthetic opioids. Despite these methodological limitations, most studies were robust, had a large sample size, and were of reasonable quality. Overall, this systematic review provides an overview of various classes and predictors of PSU patterns among individuals who use opioids, helping to inform clinical and public health interventions aimed at addressing the opioid epidemic.

Methodological issues

This review indicates that using latent class and person-centered approaches to identify distinct PSU patterns among various subgroups of individuals who use drugs is useful and informative. However, the quality assessment of the studies identified several areas for improvement for future research applying this method to drug-using populations. Most studies used unvalidated or unstandardized measurement approaches for substance use indicators. Another significant limitation of PSU analyses across most studies was the oversight of tobacco use, despite its high frequency and strong association with opioid use among individuals who use opioids. Several studies also failed to report how missing data were handled. Greater clarity and consistency regarding model fit indices and procedures for selecting the final model are essential for ensuring reproducible, reliable, and comparable results. Moreover, the notable limitations observed across studies regarding reporting and analytical approaches related to sex and gender (e.g., not reporting data on participants’ sex or gender, not conducting sex- or gender-stratified analyses, and confusing sex with gender) highlight an important knowledge gap and the need for systematically integrating and exploring sex and gender in future research on hidden PSU patterns. Some of these shortcomings have been repeatedly noted in previous reviews aimed at identifying clusters or classes of substance use and could be addressed by following existing reporting guidelines for these types of analyses.

Conclusions

This systematic review summarized evidence that applied person-centered approaches to classify PSU among individuals who use opioids. While heroin and its derivatives were the most common class, the review found that PSU was generally the norm, not the exception. However, there was considerable diversity among the PSU classes of individuals who use opioids. These findings call for further investment and research into developing treatments and interventions that extend beyond a sole focus on opioid use in this population, instead adopting a holistic and comprehensive approach to care for individuals who use opioids. Applying methodologically rigorous and transparent techniques, along with using standardized metrics for evaluating the frequency and severity of substance use patterns, is necessary to allow direct comparisons across study findings and improve their generalizability.

Introduction

Defining polysubstance use (PSU), which means using two or more different substances at once or over time, can be difficult. There is often no clear agreement on its exact definition. However, research from around the world shows that PSU is common among people who use opioids. Many drug overdose deaths also involve several drugs, not just opioids. For example, in British Columbia, Canada, fentanyl, cocaine, amphetamines, and heroin were often found in drug toxicity deaths. Similar trends are seen in the United States, where cocaine overdose deaths involving opioids increased significantly between 2000 and 2015.

Measuring PSU precisely is challenging for statisticians and researchers. Most studies have looked at the use of single drugs, which can lead to limited or biased analysis. Despite these measurement difficulties, studies consistently show that PSU is linked to a higher risk of many serious health problems.

To better understand PSU and help create programs for overdose prevention and harm reduction, researchers have developed new ways to identify groups of people with specific PSU patterns. These new methods, called "person-centered approaches," differ from older methods. Instead of focusing on how individual factors relate to outcomes, person-centered approaches assume that different groups of people in a population might respond differently to those factors. They examine how people respond to various substance use questions to find hidden types or groups of PSU patterns.

This review aims to categorize, describe, and summarize these hidden PSU patterns among people who use opioids. Identifying these patterns and what predicts them can help guide effective interventions and prevention efforts to address the ongoing overdose crisis.

Methods

Following established guidelines for systematic reviews, a search was conducted for research studies and reports on hidden classes of PSU among people who use opioids. The search covered studies published from their beginning until June 15, 2020. This included looking for search terms related to polysubstance use, opioid use, and specific analytical methods across various major scientific databases.

Studies were included if they identified hidden groups of substance use patterns among people who use opioids. This meant participants either used opioids regularly, were diagnosed with opioid use disorder (OUD) using proven methods, or were receiving treatment for OUD. Studies that included mixed groups of participants were considered if people who used opioids made up more than half of the group or were analyzed separately. Studies had to report details about the identified hidden groups and the specific substance use measures used to define them. Studies were not included if they used these analytical methods for factors other than substance use, such as HIV risk or overdose risk.

Two independent reviewers carefully screened the titles and abstracts of all identified studies. Those that met the inclusion criteria, or whose eligibility was unclear, were then fully reviewed by the same two independent reviewers. Any disagreements about including studies were resolved through discussion. Duplicate studies were removed. Data from the selected studies were then extracted into a pre-designed sheet. Due to the wide variety of factors and outcomes studied, a combined statistical analysis (meta-analysis) was not performed. Instead, the findings were summarized and described in words, focusing on the main and sub-groups of PSU identified.

The quality of the included studies and their risk of bias were assessed using a modified tool called the Newcastle-Ottawa Quality Assessment Scale. Additionally, because the review focused on specific types of hidden pattern analysis, a modified checklist called GRoLTS (Guidelines for Reporting on Latent Trajectory Studies) was used to evaluate the methodological quality of these studies, helping to ensure their transparency and how well their results could be understood.

Results

From an initial pool of 3372 unique records, 30 studies met the inclusion criteria and were included in this systematic review.

The studies varied greatly in how they defined opioid use and the types of people they included. Participants ranged from those in emergency departments, to people receiving opioid treatment, those involved with the justice system, and people accessing harm reduction services. Some studies focused only on substance use indicators, while others also included factors like age, sex, race, mental health conditions, hospital use, physical health, housing, or past overdoses. About half the studies included different measures for a single type of substance (e.g., frequency or duration of use), and half included information on how substances were administered (e.g., injection vs. non-injection). Very few studies specifically looked at fentanyl use as a unique factor.

The number of hidden PSU groups identified in each study varied widely, from two to fifteen, with four groups being the most common. While the specific groups identified differed, five main PSU patterns were found across most studies: Infrequent/low PSU; PSU mainly involving heroin; PSU mainly involving heroin and stimulants; PSU mainly involving stimulants; and Frequent PSU. Some studies also identified groups that included significant alcohol, cannabis, prescription opioids, non-opioid prescription medications, tobacco, or prescribed opioid treatment medications.

When looking at factors that predict which hidden PSU group someone belongs to, caution is needed because PSU was defined differently across studies. However, among the studies that provided detailed information, membership in higher-frequency or more intense PSU groups was often linked to an increased risk of using psychiatric services, having mental health problems, experiencing poor drug-related outcomes, frequent injecting, sharing needles, blood clots, violent behaviors, non-fatal overdose, high-risk sexual behaviors, and adverse experiences like homelessness, incarceration, unemployment, chronic pain, and hepatitis C. Only one study specifically looked at the risk of death, finding that membership in certain polysubstance injecting groups not receiving opioid treatment was linked to a higher risk of death compared to frequent buprenorphine users.

Most studies compared groups based on age, sex, and race. Increasing age was often linked to belonging to higher-frequency or more severe substance use groups, although some studies found no link. The connection between sex and PSU groups was inconsistent; females were sometimes more likely to be in groups with mental health issues or specific drug use patterns, while males were more often found in groups involving alcohol, certain drug combinations, or specific routes of administration. Findings for race and ethnicity were also inconsistent, with different racial/ethnic groups showing varied likelihoods of belonging to particular PSU classes across studies.

Overall, most included studies were of good enough quality regarding bias. While many had sufficient sample sizes and were statistically sound, most had issues with how outcomes and risk factors were measured. Assessments of hidden group studies identified several common limitations, such as using unproven or unstandardized ways to measure substance use and not always clearly reporting how missing data was handled or how the final models were chosen.

Discussion

This systematic review looked at 30 studies that used special analytical methods to identify PSU patterns among people who use opioids. Five main PSU patterns were found: infrequent/low PSU, PSU primarily involving heroin, PSU primarily involving heroin and stimulants, PSU primarily involving stimulants, and frequent PSU. The review found that using multiple substances was often the norm, not the exception, among people who use opioids. Belonging to higher-frequency or more intense PSU groups was linked to several negative mental and physical health outcomes for individuals and at a community level.

These findings are somewhat similar but not entirely comparable to previous reviews of PSU patterns in the general population or among adolescents, which often identified groups based on tobacco, alcohol, and cannabis use. These differences likely stem from the distinct ways PSU was measured and the unique substance use behaviors of people who use opioids compared to the general population or adolescents, who might be just experimenting. Future research could improve how PSU is measured and screened in clinical settings by developing and testing standardized screening tools. These tools could measure different PSU patterns, how substances are administered, reasons for PSU (e.g., recreation, withdrawal management), and recent use, which would help addiction research and clinical practice.

The review suggests that while specific PSU groups exist, using multiple substances is very common for people who use opioids. This aligns with a growing body of evidence showing that concurrent use of stimulants, prescription medications (opioid and non-opioid), and alcohol is frequent among this population. There are various reasons for PSU, including combining substances to increase euphoric effects, reducing opioid cravings or withdrawal symptoms, or balancing the effects of heroin. PSU can also be influenced by what drugs are available on the illegal market. Regardless of the reasons, PSU among people who use opioids is clearly linked to several negative mental and physical health outcomes, such as major depression, PTSD, and risky sexual practices, when compared to those who use few or no other substances.

These findings are especially important given the ongoing opioid crisis. Research and policies often focus heavily on opioids, sometimes overlooking the broader issue of multiple substance use. While supporting opioid treatment and overdose prevention is crucial, this review suggests that these interventions might have limited effectiveness for certain groups of people who use opioids, such as those who primarily use stimulants, have co-occurring alcohol use disorder, or use benzodiazepines, or those who use other substances while on opioid treatment. Recognizing these complex and diverse needs is essential for developing comprehensive care plans. Policies that do not consider these differences, or that simplify substance use by focusing only on one primary drug, may not fully address the unique risks and needs of different subgroups among people who use opioids.

Factors predicting membership in different PSU groups varied significantly across studies. While some studies found that belonging to higher-intensity PSU groups was positively linked with individual behaviors (like sharing needles, frequent injecting, high-risk sexual behaviors) and social/structural factors (like a history of incarceration, homelessness, hospitalization), findings for age, sex, and race were inconsistent. These comparisons should be viewed cautiously for several reasons. First, the way PSU was defined and the proportion of groups identified differed greatly across studies, making direct comparisons difficult. Second, most studies only looked at associations at a single point in time, which makes it hard to determine cause and effect. Lastly, many studies simply described their findings without assessing predictors, which limits the overall understanding of why people fall into certain PSU groups.

Conclusions

This systematic review summarized research that used person-centered approaches to classify PSU among people who use opioids. While heroin use was common, the review found that using multiple substances was typically the norm, not an exception. However, there was a wide variety among the specific PSU patterns identified. These findings suggest a need for further investment in research and the development of treatments and interventions that go beyond focusing solely on opioid use. Instead, care for people who use opioids should adopt a more holistic and comprehensive approach. To allow for direct comparisons across studies and improve the general applicability of findings, future research requires more rigorous and transparent methods, as well as the use of standardized ways to measure the frequency and severity of substance use patterns.

Introduction

Understanding how people use more than one substance is difficult. Polysubstance use (PSU) is a broad term. It often means using two or more different drugs at the same time or over a period. Many studies show that people who use opioids often use other drugs too. For example, in one area, most drug overdose deaths involved fentanyl, cocaine, and other drugs, not just opioids.

It is hard to measure PSU in a clear way. Many studies only look at one drug at a time. This can make it hard to get a full picture. Even so, studies agree that using many substances leads to more health problems.

Because it is so important to understand PSU, new ways of studying it have been developed. These new ways look at groups of people and their drug use patterns, instead of just looking at single drugs. This helps researchers find different types of drug use patterns among people.

This review looks at these patterns of PSU in people who use opioids. The goal is to describe these patterns. Finding these patterns helps create better programs to prevent overdose and reduce harm.

Methods

This review followed specific guidelines to find studies about polysubstance use (PSU) among people who use opioids (PWUO). Researchers searched several databases up to June 2020. Studies were included if they looked at different groups or patterns of substance use in people who regularly used opioids, had an opioid use disorder, or were getting treatment for it. Studies with mixed groups were included if more than half the people used opioids.

Researchers only included studies that found specific patterns of drug use using special methods. They checked that these studies explained the drug use patterns clearly. Studies were not included if they used these methods for other health issues like HIV risk. Two different people checked all studies to decide if they should be included. If they disagreed, they talked it over.

Information was collected from each study about who was in it and what patterns of drug use were found. It was hard to compare all the studies directly because they used different ways to measure drug use and looked at different things. So, the findings were described in words instead of combining numbers from all studies. The quality of each study was also checked to see how reliable the information was.

Results

Researchers looked at 3,372 studies and included 30 of them in this review. All studies focused on people who used opioids, but how they defined "opioid use" was different. The people in the studies also varied a lot. They included people in emergency rooms, people getting treatment for opioid use, and people who used drugs and were involved with the justice system.

Most studies mainly looked at drug use. Some also looked at things like age, sex, race, or mental health. They found between 2 and 15 different drug use patterns in the studies, with 4 being the most common. Even though the patterns varied, five main types of polysubstance use (PSU) were found across many studies:

• Not using many substances or using them rarely.

• Mostly using heroin.

• Mostly using heroin and stimulants.

• Mostly using stimulants.

• Using many substances often.

The studies compared these drug use patterns with other traits to see what might predict them. It was found that older age was often linked to using more substances, but this was not always true. What linked to sex (male or female) was not clear, as findings were different across studies. For race or ethnicity, the results were also mixed and not consistent across studies.

For people getting treatment for opioid use (OAT), studies showed different results. Some found that people on OAT were less likely to use many substances. Others showed that even on OAT, people might use other drugs, but often less intensely than those not on OAT. It was also noted that some people used non-prescribed buprenorphine to manage withdrawal, not to get high.

Being in a group that used more substances was linked to worse health outcomes. These included needing more mental health services, having mental health problems, having more drug-related problems, injecting drugs often, sharing needles, having blood clots, being violent, having non-fatal overdoses, and engaging in risky sexual behaviors. It was also linked to problems like homelessness, being in jail, not having a job, chronic pain, and having Hepatitis C. One study found a higher risk of death for people who injected many substances and were not getting OAT. Most studies were of good quality, but some had problems with how they measured drug use.

Discussion

This review looked at 30 studies about how people who use opioids also use other drugs. It found five main patterns of using multiple substances: infrequent use, mostly heroin, heroin and stimulants, mostly stimulants, and frequent use of many drugs. This suggests that using multiple drugs is common, not rare, among people who use opioids. Those who used more drugs had more health problems.

These findings are different from reviews of polysubstance use in the general public or teenagers, likely because the reasons for drug use and patterns are different in people who use opioids. It suggests that doctors and researchers need better ways to ask about and understand all the different drugs people are using. New tools could help identify people at higher risk and create better, more personal treatment plans that go beyond just focusing on one drug.

People use multiple drugs for many reasons. Sometimes it's to make the drug effects stronger, or to balance them out. Other times, it's to reduce withdrawal symptoms or because certain drugs are available on the street. No matter the reason, using many substances is linked to serious health issues like mental health problems, risky behaviors, and even overdose. These findings are very important because many programs fighting the opioid crisis focus only on opioids. They might miss the fact that many people who use opioids also use other drugs.

This means that programs should not just focus on opioids. They need to look at all the drugs someone is using and their full life situation. Treating only one drug use problem might not be enough for some people. Programs need to be flexible and meet the unique needs of different groups of people who use opioids, especially those who also use stimulants, alcohol, or other prescription drugs.

However, this review has some limits. The studies included different groups of people and measured drug use in different ways, making direct comparisons hard. Also, most people in the studies were male, White, and from North America, so the findings might not apply to everyone. The drug supply is also changing quickly, with new strong opioids appearing, which can change drug use patterns. Future studies need to consider how these changes affect polysubstance use. Also, many studies did not fully include tobacco use in their analyses, even though it is common among people who use opioids. There is a need for clearer, more consistent methods in future research, including better ways to report on study details and how different genders are affected.

Conclusions

This review shows that people who use opioids often use other substances as well. While heroin use was common, using many different drugs was the usual pattern, not the exception. The ways people used these multiple drugs varied greatly. This means that treatments and programs should not just focus on opioids. They need to use a wider approach to help people who use opioids by considering all the substances they use. Future research needs to use clear and standard ways to measure drug use patterns so that findings can be better compared and understood.