Characterizing the clinical subgroups of individuals who present to the emergency department for alcohol-related harms in Ontario, Canada: A latent class analysis

Alcohol-related emergency department (ED) visits are common and associated with adverse clinical outcomes, including premature mortality. This population-based retrospective cohort study identified clinically distinct subgroups of individuals who experience alcohol-related ED visits and characterized differences in the risk of adverse outcomes between them. 73,658 individuals who experienced an alcohol-related ED visit in Ontario, Canada between 2017 and 2018 were identified. Latent class analysis (LCA) revealed five clinically distinct subgroups within the overall cohort. These subgroups followed a severity gradient from low-frequency service use for acute intoxication to high-frequency service use for alcohol use disorder (AUD) and related comorbidities. Relative to those presenting for acute intoxication, those presenting for AUD and comorbidities had a much higher risk of hospital admission (adjusted odds ratio [aOR]: 8.26, 95 % confidence interval [CI]: 7.81 – 8.75) and post-discharge mortality (adjusted hazard ratio [aHR]: 3.07, 95 % CI: 2.81 – 3.37). There was a subgroup of individuals with a history of high


Introduction
Alcohol-related emergency department (ED) visits are common and incur a large cost to healthcare systems around the world (Butler et al., 2016;Friesen et al., 2022b;Myran et al., 2021;Myran et al., 2019b;Sherk, 2020;Verelst et al., 2012;White et al., 2018).Recent data from Canada and the United States (US) has indicated that rates of alcohol-related ED visits have increased substantially over the past two decades and that individuals who experience recurrent alcohol-related ED visits are at a high-risk of premature mortality (Hulme et al., 2020;Myran et al., 2019b;White et al., 2018).However, not all individuals who experience alcohol-related ED visits are alike, and there are a wide variety of mechanisms through which alcohol use necessitates emergency care.For example, some alcohol-related ED visits are related to acute intoxication, which may or may not be the primary diagnosis (e.g., alcohol poisoning versus an injury that occurred while intoxicated), whereas others can be related to the health sequelae associated with chronic high-volume alcohol use, such as withdrawal, psychosis, pancreatitis, or liver disease (Canadian Institute for Health Information, 2017).
The clinical diversity that exists in this cohort likely translates into different patterns of health service use and risk of adverse outcomes between the different subpopulations of individuals who experience alcohol-related ED visits.Therefore, it is prudent to understand the unique epidemiological trends associated with each of these different clinical populations.To date, however, most population-based research on this topic makes use of composite indictors (i.e., broad sets of diagnostic codes) to identify ED visits that are attributable to alcohol, resulting in an amalgamation of all those who experience alcoholrelated ED visits into a single cohort (Canadian Institute for Health Information, 2020;Friesen et al., 2022b;Hulme et al., 2020;Myran et al., 2019a).While this has clarified overall trends in alcohol-related ED visits, it has failed to delineate the clinically distinct subgroups of individuals who present to the emergency department for alcohol-related harms.This, in turn, limits our capacity to identify individuals at the highest risk of adverse outcomes and create tailored strategies to mitigate the risk of downstream harm.
The purpose of this project was to address this gap in knowledge by evaluating if there are clinically distinct subgroups of individuals who experience alcohol-related ED visits based on (1) the presenting diagnosis and (2) the history of prior alcohol-related health service use.We then analyzed whether there were differences between subgroups in terms of sociodemographic composition and risk of adverse outcomes.

Study design
This was a population-based retrospective cohort study of individuals who experienced an alcohol-related ED visit in Ontario, Canada between January 1, 2017, andDecember 31, 2018 (the 'index ED visit').Ontario is Canada's most populous province and collects health administrative data for all ED visits that occur in the province, which provided an opportunity to conduct this analysis using a comprehensive, population-based sample.Alcohol-related ED visits were identified in the National Ambulatory Care Reporting System (NACRS) database using a previously described indicator developed by the Canadian Institute for Health Information (CIHI) that captured ED visits that are entirely attributable to alcohol use (CIHI, 2020).This indicator is commonly used in Canadian epidemiological research on alcohol-related harms and parallels the indicator developed by the Centers for Disease Control and Prevention (CDC) that is commonly used in the United States (US) (Alcohol-related Disease Impact Scientific Working Group, 2016).
Individuals were excluded from the cohort if they (1) were aged <10 or >105, (2) were not Ontario residents, (3) were not eligible for universal healthcare coverage for the entirety of the study period, (4) had an invalid ICES key number (required for databased linkage), or (5) died in the ED during the index event.If an individual had more than one alcohol-related ED visit during the accrual window, one was selected at random to be the index event, as has been done previously (Friesen et al., 2023;Hulme et al., 2020).After the cohort was created, we identified (1) a one-year follow up window from the date of discharge from the index ED visit to collect data on study outcomes and (2) a two-year lookback window from the date of the index ED visit to collect data on previous alcohol-related health service use.
Linked health administrative databases housed at ICES, Ontario's largest health data repository, were used to identify all data used for this study (see Supplementary Table 1 for the complete list of databases).ICES databases contain health service use information for all Ontarians with universal healthcare coverage through the Ontario Health Insurance Plan (OHIP), which represents virtually all residents of the province.The use of the data in this project is authorized under section 45 of Ontario's Personal Health Information Protection Act (PHIPA) and does not require review by a Research Ethics Board.

Exposures
The exposures of interest were (1) the alcohol-related diagnostic code(s) associated with the index ED visit (multiple codes can be listed within a single record) and the count of (2) alcohol-related outpatient visits, (3) alcohol-related ED visits, and (4) alcohol-related hospitalizations in the two years prior to the index visit (see Supplementary Table 1 for the databases used to capture these variables).These exposures were then used to conduct a latent class analysis (LCA) to identify clinical subgroups of individuals in the cohort based on their presenting health condition(s) and their history of alcohol-related health service use.For the purposes of the LCA, the alcohol-related diagnostic codes associated with the index ED visit were operationalized as binary variables whereby an individual either had a 0 or a 1 assigned for each diagnostic code outlined in the CIHI indicator (see Table 1).The count of prior alcohol-related health service encounters was operationalized as a categorical variable whereby an individual had either 0, 1, or 2+ previous outpatient visits, ED visits, or hospitalizations in the 2 years prior to the index ED visit.The subgroup that an individual was assigned to by the LCA subsequently became the exposure of interest for the multivariable regression models outlined below.

Outcomes
The outcomes of interest were (1) admission to hospital from the index ED visit, (2) time to repeat alcohol-related ED visit within 1-year of discharge from the index visit, and (3) time to all-cause mortality within 1-year of discharge from the index visit (including time to inhospital mortality if the individual was admitted to hospital).The databases used to capture each of these outcomes are outlined in Supplementary Table 1.

Covariates
Covariates that could foreseeably confound the association between exposures and outcomes were chosen a priori based on previous work on alcohol-related health service use in Canada (Friesen et al., 2022a(Friesen et al., , 2023;;Myran et al., 2021aMyran et al., , 2021bMyran et al., , 2019b;;Nickel et al., 2018).These variables included age, sex, income quintile (measured using after-tax income from the 2016 Canadian census), rurality (measured using Statistical Area Classification (Statistics Canada, 2016)), medical comorbidity (measured using the Johns Hopkins ACG® System Aggregated Diagnosis Groups [ADG] Version 10 score), and psychiatric comorbidity (measured using a previously described psychiatric severity gradient based on psychiatric service use [outpatient, emergency, and inpatient] during the two-year lookback (Klaassen et al., 2019)).The databases used to capture each of these covariates are outlined in Supplementary Table 1.

Statistical analysis
The four exposure variables were input into an LCA, which was run using the poLCA package in R version 4.0.0.LCA models with increasing numbers of classes were created and the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC) were tabulated for each model.The final model specification was chosen based on (1) the point at which the BIC no longer meaningfully improved when an additional class was added, and (2) the clinical interpretability of the subgroups, as has been done previously (Hastings et al., 2014).Individuals were assigned to the class to which they had the highest calculated probability of membership.To understand the clinical characteristics of the identified classes, descriptive statistics of the exposure variables (presenting diagnosis and count of previous alcohol-related health service encounters) were tabulated by subgroup.Descriptive statistics of the sociodemographic and clinical covariates were also tabulated after stratifying the cohort by subgroup.Significant differences between subgroups were gauged using Chi square tests for independence (categorical variables) and one-way analysis of variance (ANOVA, continuous variables).
Next, we evaluated the association between subgroup membership and the study outcomes.The association between subgroup membership and hospital admission (binary variable) was assessed using a multivariable logistic regression model.The association between subgroup membership and time to repeat alcohol-related ED was evaluated using a multivariable Fine & grey subdistribution hazard model with all-cause mortality included as a competing risk.The association between subgroup membership and time to death was evaluated using a multivariable Cox proportional hazards model.Each model was adjusted for all covariates listed above.~1 % of the cohort was missing data for Statistical Area Classification (measure of rurality) and income (see Table 1), and these individuals were not included in the statistical models.For completeness, unadjusted models were also run, and the results were presented alongside the adjusted models.Apart from the LCA, all statistical analyses were run in SAS version 9.4.

Table 1
Index diagnoses and prior alcohol-related health service use across the five subgroups identified in the latent class analysis.
For clarity, shading is representative of the proportion of each subgroup with that characteristic (darker shade, higher proportion).* Numbers represent the % of each subgroup with that characteristic.Columns do not sum to 100 % because individuals can have multiple diagnoses associated with the index hospitalization.

Results from the LCA
After the exclusion criteria were applied (Supplementary Table 2) a total of 73,658 individuals were included in the cohort.The LCA identified that a model with five classes was the preferred specification (Supplementary Table 3).For conceptual clarity, these five classes and the proportion of the cohort falling into each are visualized in Fig. 1.The specific clinical characteristics of the five classes are also detailed in Table 1.Based on these characteristics, the classes were characterized as follows: (1) individuals presenting with acute intoxication and the lowest frequency of prior alcohol-related health service use (44.3 % of cohort), (2) individuals presenting for alcohol poisoning and a low-average frequency of prior alcohol-related health service use (4.1 % of cohort), (3) individuals presenting for harmful alcohol use and an average frequency of prior alcohol-related health service use (21.4 % of cohort), (4) individuals presenting for alcohol dependence, withdrawal, and alcoholrelated organ dysfunction (termed alcohol use disorder (AUD) with comorbidities) and a high-average frequency of prior alcohol-related health service use (22.9 % of cohort), and (5) individuals presenting for a variety of diagnoses and the highest frequency prior alcohol-related health service use (7.3 % of cohort; Table 1, Fig. 1).

Sociodemographic and clinical differences between subgroups
Descriptive statistics of the sociodemographic and clinical covariates, stratified by subgroup, are presented in Table 2. Overall, the average age of the cohort was 40.89 years (standard deviation [SD]: 18.22), 36.85 % were female, 31.78 % were in the lowest income quintile (versus 14.76 % in the highest income quintile), 76.71 % lived in a large metropolitan area, the average ADG (medical comorbidity) score was 14.04 (standard deviation [SD]: 14.84), and 56.58 % had some form of prior non-alcohol-related psychiatric service use.
Relative to these overall trends, individuals in the acute intoxication subgroup were younger (average age: 36.66 years, SD: 18.32), proportionately more female (40.36 % female), and had a lower rate of medical comorbidities (average ADG score: 9.96, SD: 13.21) and psychiatric comorbidities (50.06 % had prior psychiatric service use).Individuals in the alcohol poisoning subgroup were also younger (average age: 35.43,SD: 16.34) and proportionately more female (52.63 % female), but had high rates of psychiatric comorbidity (70.46 % had prior psychiatric service use).Individuals in the AUD with comorbidities subgroup were proportionately older (average age: 48.04, SD: 16.73), more male (29.51 % female), higher income (20.97% in the lowest income quintile), and had a higher burden of medical comorbidities (average ADG score: 18.42, SD: 15.28).Individuals in the frequent service use category were also proportionately older (average age: 45.63, SD: 13.80), more male (31.53 % female), and had high burden of medical comorbidities (average ADG score: 26.03, SD: 15.09); however, these individuals were distinct in that they were also more likely to live in a low income neighborhood (39.79 % in the lowest income quintile) and had high rates of psychiatric comorbidity (85.70 % had prior psychiatric service use).

Differences in the risk of adverse outcomes between subgroups
The cumulative incidence of each study outcome is presented in Fig. 2. Overall, 20.02 % of individuals were admitted to hospital from the ED, 20.33 % had at least one additional alcohol-related ED visit in the year following the index visit, and 4.95 % of individuals died within one year of discharge from the index visit.There were clear differences in the incidence of each outcome between subgroups.Individuals in the alcohol poisoning and AUD with comorbidities subgroups had the highest frequency of hospital admission (35.96 % and 42.79 %, respectively).Individuals in the frequent service use subgroup had the highest incidence of repeat ED visit in the year following discharge from the index event (68.35 %).Individuals in the AUD with comorbidities and frequent service use subgroups had the highest incidence of all-cause mortality in the year following discharge from the index event (10.42 % and 8.86 %, respectively).

Discussion
In this population-based retrospective cohort study, we identified five clinically distinct subgroups of individuals who experienced alcohol-related ED visits in Ontario, Canada.These subgroups followed a severity gradient that ranged from individuals presenting for acute intoxication to individuals presenting for advanced AUD and alcoholrelated organ dysfunction.There were clear differences in the frequency of alcohol-related health service use between groups, with the defining feature of one subgroup being a disproportionately high rate of alcohol-related ED use both prior to and after the index visit.Each subgroup also had a unique sociodemographic composition and differential risk of adverse outcomes.Notably, there was (1) a subgroup of younger individuals that presented for alcohol poisoning, had high rates of psychiatric comorbidity, and were at the highest risk of being admitted to hospital from the ED, and (2) a subgroup of older individuals who presented for alcohol dependence, withdrawal, and organ dysfunction and were at the highest risk of risk of 1-year mortality.These findings provide important context about the clinical subgroups of individuals who experience alcohol-related ED visits and suggest that there are certain subgroups that could merit prioritization in efforts to reduce the high rates of recurrent health service use and mortality observed in this cohort.
In the past decade, there has been a substantial amount of research  done on alcohol-related ED visits in Ontario and Canada.Notable findings include (1) that rates of alcohol-related ED visits in Ontario increased by 83 % between 2003 and 2016 and rose disproportionately among certain demographic subgroups (e.g., individuals with lower income, females, rural residents) (Myran et al., 2021a(Myran et al., , 2019b)), ( 2) that the proportion of ED visits attributable to alcohol increased during the COVID-19 pandemic (Myran et al., 2021b), (3) that rates of alcohol-related ED visits vary dramatically between geographic regions and are particularly high in rural, Northern communities (Friesen et al., 2022b), and (4) that a higher frequency of alcohol-related ED use is associated with a higher risk of short-term mortality (Hulme et al., 2020).This Canadian work is paralleled by international studies, such as a US report that found that national rates of alcohol-related ED visits increased by 47 % between 2006 and 2014 (White et al., 2018), and a Californian study that found that the injury-related mortality rate was 15-fold higher among individuals who experienced an alcohol-related ED visit than among demographically matched controls (Goldman-Mellor et al., 2022).
Together, this research has highlighted that international rates of alcohol-related ED visits are high, increasing over time, and associated with significant morbidity and mortality.This, in turn, has laid the foundation for research into primary prevention strategies to reduce rates of alcohol-related ED visits in the general population (e.g., reducing alcohol outlet density (Gruenewald, 2011), warning labels on alcohol packaging (Hobin et al., 2020)), as well as secondary prevention strategies to reduce the risk of recurrent harm and death among  individuals who experience alcohol-related ED visits (e.g., Addiction Consult Services (Weinstein et al., 2018), Rapid Access Addiction Medicine (RAAM) clinics (Corace et al., 2020)).Importantly, however, most of this epidemiological and preventative healthcare research amalgamates all individuals who experience an alcohol-related ED visit into a single group.As was made clear in this study, this is an oversimplification of a much more complex reality.The distinct clinical subpopulations that were identified suggest that strategies to reduce the risk of in-hospital or post-discharge harm may need to be tailored to the unique needs of each subgroup, rather than applied broadly to the entire cohort.For example, it is unlikely that the same clinical approach will be equally effective in reducing downstream alcohol-related harm for those in the alcohol poisoning subgroup versus those in the AUD with comorbidities subgroup, given the substantial demographic and clinical differences that exist between them.Furthermore, certain subgroups may merit prioritization in our efforts to reduce the risk of post-discharge adverse outcomes.For example, those who experience high-frequency alcohol-related ED visits may be a reasonable target for strategies aimed at addressing recurrent alcohol-related health service use, given that they were at a disproportionately high risk of a repeat ED visit.Additional research is required to understand how our clinical strategies can be tailored to the unique needs of the clinical subgroups identified in this study.
This study also provides important insight on the indicators that are used to identify alcohol-related ED visits in health administrative databases.In general, these indicators are broad and contain diagnostic codes for the many health conditions associated with high-volume and/ or chronic alcohol use (Rehm, 2011).For example, the CIHI indicator used in this study contains 52 ICD-10 and 10 Diagnostic and Statistical Manual -5th Edition [DSM-V] diagnostic codes (Canadian Institute for Health Information, 2020).As described above, broadly applying these composite indicators to capture alcohol-related ED visits misses a degree of underlying variable clinical complexity.At the same time, however, it may not be necessary or practical to consider each diagnostic code separately when accounting for this complexity in descriptive or statistical analyses.Rather, our LCA placed these codes into clinically meaningful groups that we have labeled 'acute intoxication', 'alcohol poisoning', 'harmful use', and 'AUD with comorbidities'.In turn, these groupings may warrant consideration in future work that makes use of similar indictors to identify and study alcohol-related ED visits using health administrative databases.

Limitations
This study has several limitations.First, this study was conducted in a single Canadian province and patterns of alcohol use, alcohol-related harms, and alcohol-related health service are known to vary regionally (Friesen et al., 2021;Griswold et al., 2018;Canadian Institute for Health Information, 2017).The subgroups of individuals that experience alcohol-related ED visits may vary between Canadian and international jurisdictions, limiting the external validity of the findings.Nonetheless, this study used a comprehensive, population-based sample of ED visits in a province of ~15 million inhabitants, which supports the generalizability of the findings to populations with similar sociodemographic composition and healthcare systems.Second, this study identified a severity gradient in the subtypes of individuals experiencing alcohol-related ED visits that ranged from low-frequency service use for acute intoxication to high-frequency service use for severe AUD and related comorbidities.Due to the study design, we could not track how individuals progressed between these categories over time.This will be an important avenue of future investigation, as it will allow for us to understand whether and how individuals progress along this severity gradient, such that targeted strategies can be created to modify this progression and reduce the risk of adverse outcomes.Third, due to limitations associated with the health administrative data used in this study, it was not possible to control for all factors that could confound the association between subgroup membership and adverse outcomes.Factors such as polysubstance use, stigmatization, help-seeking behaviors, and the relative accessibility of mental health services (just to name a few) can influence the ways in which people with AUD access health services and their risk of downstream harm.In turn, these factors will need to be considered in future work, which will likely require the use of both qualitative and quantitative designs.Finally, due to data limitations, this study only captured physician-based alcohol-related health service use.This is only a portion of the types of care that individuals with AUD receive, which can also include private addiction treatment services and mental health care for comorbid psychiatric conditions that contribute to an AUD (Friesen et al., 2023).Considering these additional health services in future work will be important to capture a complete picture of the clinical trajectories of individuals who experience alcohol-related ED visits.

Conclusion
This population-based retrospective study identified five distinct clinical subgroups of individuals who experienced alcohol-related ED visits in Ontario, Canada.Each subgroup had unique sociodemographic characteristics and risk of adverse outcomes, illustrating the clinical diversity that exists within this cohort.Notably, the subgroups study followed a severity gradient whereby a few small subgroups, particularly those with frequent alcohol-related health service use and severe AUD, had a disproportionately high risk of adverse outcomes.This suggests that there is an identifiable subset of individuals who present to the ED for alcohol-related harms that may require prioritization in preventative efforts to reduce the high rates of readmission and mortality seen in the overall cohort.These results also have implications for future epidemiological work on alcohol-related ED visits, insofar as the identified subgroups may warrant consideration during both study design and data interpretation.

Declaration of competing interest
Nothing to declare.
E.L.Friesen et al.

Fig. 1 .
Fig. 1. Results from the Latent Class Analysis (LCA).In a cohort of individuals who experienced an alcohol-related ED visit between January 1, 2017, and December 31, 2018 (the 'index event'), 5 distinct subgroups of individuals were identified by the LCA.Characteristics of the subgroups in term of index diagnosis and history of alcohol-related health service use are presented in Table1.Percentages represent the proportion of each cohort that belonged to each subgroup.

Fig. 2 .
Fig. 2. Cumulative incidence of study outcomes in the Ontario cohort, stratified by subgroup.(A) Cumulative incidence (readmission) and hazard (all-cause mortality) curves for outcomes in the year following discharge from the index hospitalization, stratified by subgroup.(B) 1-year incidence of each study outcome, stratified by subgroup.

Table 2
Demographic and clinical characteristics of the cohort, stratified by subgroup.

Table 3
Results from the multivariable regression models for in-hospital mortality, readmission, and post-discharge mortality.