Outpatient psychiatric service use is associated with a reduced risk of 1‐year readmission and mortality following alcohol‐related hospitalizations: A historical cohort study

Alcohol‐related hospitalizations are common and associated with high rates of short‐term readmission and mortality. Providing rapid access to physician‐based mental health and addiction (MHA) services post‐discharge may help to reduce the risk of adverse outcomes in this population. This study used population‐based data to evaluate the prevalence of outpatient MHA service use following alcohol‐related hospitalizations and its association with downstream harms.

K E Y W O R D S alcohol use disorder, alcohol-related hospitalization, mental health, mortality, readmission

| INTRODUCTION
Alcohol use is a leading risk factor for morbidity and premature mortality worldwide. 1 In Canada, there are more than 80,000 hospital stays caused by the acute and chronic effects of alcohol use every year, which incurs a large burden on the healthcare system. 2 Almost all individuals who are hospitalized for an alcohol-related health condition have a current alcohol use disorder (AUD) or a recent history of AUD and are at a high-risk of recurrent alcohol-related harm following discharge. 3,4 For example, recent studies have estimated that 20%-60% of individuals are readmitted to hospital and 10%-40% of individuals die in the year following discharge from an alcohol-related hospitalization, with variability based on the underlying health condition and follow-up time. [4][5][6][7][8][9] Given the high prevalence of alcohol-related hospitalizations, strategies to break the cycle of recurrent harm in this population are urgently required.
Facilitating rapid access to AUD treatment following alcohol-related hospitalizations is an effective strategy for reducing the likelihood of recurrent harm in this highrisk population. For example, Peeraphatdit et al. found that initiation of an alcohol rehabilitation program (including inpatient, outpatient, or peer support programs) within 30 days of discharge from a hospitalization for alcoholic hepatitis was associated with an $80% reduction in the odds of 30-day readmission and mortality in an American cohort. 4 In the Canadian context, the use of rapid access addiction medicine (RAAM) clinics following alcohol-related health service encounters has been associated with a reduced risk of recurrent alcoholrelated emergency department (ED) visits. 10 However, many individuals who are hospitalized for alcohol-related health conditions have a comorbid psychiatric illness, which, if present, is associated with an increased risk of short-term harm post-discharge. 9 In turn, providing access to a broader suite of mental health services following alcohol-related hospitalizations (i.e., beyond AUD treatment) may help to further reduce the risk of recurrent harm in this population.
To date, no studies have used population-based data to evaluate how often individuals use physician-based outpatient mental health and addiction (MHA) services following alcohol-related hospitalizations, or whether receiving such services is associated with a reduced risk of recurrent harm post-discharge. Ontario, Canada collects comprehensive data on health service use (including outpatient, emergency, and inpatient care) for its 14 million residents, which provides a unique opportunity to track population-level trends in health service use and clinical outcomes. This study leveraged this data to evaluate (1) the prevalence of outpatient MHA service use (from either a psychiatrist or primary care physician) within 30 days of discharge from an alcohol-related hospitalization, and (2) the association between receiving outpatient MHA services and the hazard of 1-year readmission and mortality.

| Study design
This was a population-based historical cohort study of individuals living in Ontario, Canada who had an alcohol-related hospital discharge between January 1, 2016 and December 31, 2018. Alcohol-related

Significant outcomes
• Short-term outcomes following alcohol-related hospitalizations were poor -19% of individuals were readmitted to hospital and 12% of individuals died within 1 year of discharge. • Receiving outpatient mental health and addiction (MHA) services following alcohol-related hospitalizations was associated with a reduced hazard of 1-year readmission and mortality. • Only 1 in 5 individuals received MHA care in the 30 days following an alcohol-related hospitalization.

Limitations
• Despite adjusting for relevant confounding variables, unmeasured differences between individuals who did and did not receive MHA services may have influenced the observed associations. • Only physician-based MHA services were captured in this study, which does not include MHA care provided by psychologists or at private addiction treatment facilities.
hospitalizations were identified within two hospital discharge databases, the Canadian Institute for Health Information (CIHI) Discharge Abstract Database and the Ontario Mental Health Reporting System, which capture virtually all hospital stays in Ontario. Alcohol-related hospitalizations were defined using a previously described indicator, "Hospitalizations Entirely Caused by Alcohol Use," that uses International Classification of Disease-10th Edition (ICD-10) and Diagnostic and Statistical Manual of Mental Disorders (DSM-5) codes to identify alcohol-related hospitalizations in the abovementioned databases (Table S1). For individuals with more than one alcohol-related hospitalization in the accrual window, one was chosen at random to be the "index hospitalization." Individuals were excluded from the cohort if the following criteria were met: (1) age <10 or >105 years, (2) non-Ontario resident, (3) invalid postal code, (4) invalid ICES Key Number (IKN, required for database linkage), (5) ineligibility for provincial healthcare coverage (Ontario Health Insurance Plan [OHIP]) at any point during the study timeframe, and (5) death prior to discharge from the index hospitalization. In addition, individuals who were readmitted to hospital or died within 30 days of discharge from the index hospitalization were excluded to mitigate the immortal time bias associated with the study exposure (i.e., to facilitate the landmark analysis described below). The dataset used in this study was derived from one that was used in a previous analysis of rural-urban disparities in alcohol-related health service use in Ontario. 11 Linked health administrative databases housed at ICES, Ontario's largest health data repository, were used to identify all data. ICES databases contain health service use information for Ontarians with healthcare coverage through OHIP, which represents virtually all residents. 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.

| Exposure
The primary exposure was the use of outpatient MHA services in the 30 days following discharge from the index alcohol-related hospitalization. This was defined as either a psychiatrist visit or MHA-related primary care visit (assessed as separate exposures and as a composite). These events were identified in the OHIP Claims Database, which contains data on outpatient health service use in Ontario. MHA-related primary care visits were defined using a previously described algorithm based on the ICD-9 code associated with the primary care visit. The 30-day exposure window for MHA service use was chosen to align with an existing report that receipt of alcohol rehabilitation services in the first 30 days of discharge from an alcohol-related hospitalization is associated with a reduced risk of short-term adverse outcomes. 4 This "landmark analysis" design also served to mitigate immortal time bias, as all cohort members had the opportunity to experience the exposure because individuals who died or were readmitted to hospital in the first 30 days were excluded from the cohort. We opted to conduct a landmark analysis rather than evaluating MHA service use as a time-dependent covariate (another option to control for immortal time bias) given the challenges associated with estimating the effect size of an internal time-dependent covariate in survival analyses with competing risks. 12

| Outcomes
There were two outcomes of interest, (1) alcohol-related hospital readmission and (2) all-cause mortality in the year following discharge from the index alcohol-related hospitalization. The former was defined using the same criteria used to identify the index alcohol-related hospitalization. The latter was determined using the Registered Persons Database, which provides information on date of death. The time(s) between discharge from the index hospitalization and (1) the first readmission and (2) death were calculated for use in the time-to-event regression models described below. Outcomes occurring beyond 1 year of discharge from the index hospitalization were not captured because the purpose of this study (and the previous work that made use of this dataset 11,13 ) was to investigate the association between MHA service use and short-term outcomes following alcohol-related hospitalizations.

| Covariates
Covariates were chosen to be included in the statistical models a priori if they could foreseeably confound the association between outpatient MHA service use and the study outcomes based on clinical rationale and a literature review. 4,9,11,14,15 These variables included age, sex, income quintile, length of index hospitalization (in days), type of index hospitalization (see Table S1), rural versus urban residence (defined as a Rurality Index of Ontario score ≥40), medical comorbidity burden (measured using The Johns Hopkins ACG ® System Aggregated Diagnosis Groups (ADG) Version 10 score), psychiatric comorbidity burden (measured using a previously described "psychiatric utilization gradient"), and the number of alcohol-related hospitalizations in the 2 years prior to the index hospitalization. The psychiatric utilization gradient is a categorical scale of mental illness severity that is based on psychiatric service use in the 2 years prior to the index hospitalization (categories: none, outpatient, emergency, hospitalization). 16

| Statistical analysis
Descriptive statistics of all covariates, stratified by whether an individual received outpatient MHA care in the 30 days following discharge from the index alcoholrelated hospitalization, were tabulated. Significant differences between groups were gauged using standardized difference (SD), where an SD ≥0.10 indicated a significant difference between groups. 17 The association between post-discharge MHA service use and time to alcohol-related hospital readmission was evaluated using multivariable Fine and Grey subdistribution hazard model with all-cause mortality as a competing risk. The association between post-discharge MHA service use and all-cause mortality was evaluated using a multivariable Cox proportional hazards model. In both cases, the results from the unadjusted and adjusted models were tabulated. Unadjusted cumulative incidence and cumulative hazard plots were generated from the subdistribution and proportional hazards models, respectively, to visualize the unadjusted association between receipt of MHA services and the study outcomes over time.
We conducted three additional sensitivity analyses. First, the use of landmark analysis resulted in the exclusion of individuals who died or were readmitted to hospital within the first 30-days of discharge from the index hospitalization. In turn, the results of this study cannot generalize to this subset of high-risk individuals experiencing early post-discharge harm. To better understand who these individuals are, and how they differ from those who were retained in the cohort (i.e., did not experience an adverse outcome in the first 30 days post discharge), we tabulated their demographic and clinical characteristics and compared them to those of the primary cohort using standardized differences. Second, we re-ran the multivariable models after stratifying the exposure (receipt of outpatient MHA services) by the type of service provider, that is, psychiatrist or primary care physician, to evaluate if there were provider-specific associations with readmission and mortality. Third, we conducted a mediation analysis to evaluate a priori hypothesis that receiving post-discharge, outpatient MHA services would reduce the likelihood of readmission (by addressing AUD and comorbid mental illness), which, in turn, would reduce the risk of 1-year mortality. This was done using the counterfactual approach described by Lange et al., whereby marginal structural models were used to break down the total effect (TE) of MHA care on all-cause mortality into the natural direct effect (NDE) of MHA care on mortality and the natural indirect effect (NIE) of MHA care on mortality, mediated by the effect of MHA care on readmission (see Figure S1). 18 These marginal structural models were adjusted for all study covariates and 95% confidence intervals were generated via bootstrapping (n = 500), as done previously. 11,19 All statistical analyses were conducted in SAS version 9.4. 20

| Descriptive statistics
A total of 43,343 individuals were included in the cohort after the exclusion criteria were applied (see Table S2 for a count of exclusions by reason). 8589 (19.8%) of these individuals received outpatient MHA services in the 30 days following discharge from the index alcoholrelated hospitalization (Table 1). Individuals who received outpatient MHA services, versus those who did not, were younger (mean age: 45.3 years vs. 53.6 years, SD: 0.50), proportionately more female (37.4% vs. 29.1%, SD: 0.18), more likely to have a psychiatric comorbidity (prior inpatient psychiatric service use: 36.7% vs. 25.2%, SD: 0.25), and more likely to have had their index alcohol-related hospitalization attributed to AUD (57.0% vs. 41.5%, SD: 0.31) than to other diagnoses such as harmful alcohol use (9.8% vs. 16.7%, SD: 0.21) or alcoholic liver disease (2.7% vs. 10.0%, SD: 0.31; Table 1). Income, medical comorbidity burden, length of the index hospitalization, rural-urban status, and number of previous alcohol-related hospitalizations were not significantly different between those who did and did not receive outpatient MHA services following the index hospitalization (Table 1).

| Receipt of outpatient MHA services is associated with a reduced hazard of readmission and mortality
Overall, 14.8% of the cohort were readmitted to hospital and 8.4% of the cohort died in the year following the index alcohol-related hospitalization (excluding those who died or were readmitted to hospital in the first 30 days following discharge). The results from the unadjusted and adjusted models for the associations between receiving outpatient MHA care and (1) readmission and (2) mortality are presented in Table 2. The unadjusted cumulative incidence of alcohol-related readmission and cumulative hazard of all-cause mortality, stratified by receipt of MHA services, is presented in Figure 1. This illustrated that receiving MHA services was associated with a 7% reduced hazard of readmission (hazard ratio [HR] 0.93, 95% confidence interval [CI]: 0.87-0.99) and a 54% reduced hazard of all-cause mortality (HR: 0.46, 95% CI: 0.42-0.51). After adjusting for all study covariates, receiving any outpatient MHA services was associated with a 6% reduced hazard of readmission (adjusted hazard ratio [aHR] 0.93, 95% CI: 0.88-0.99) and a 26% reduced hazard of all-cause mortality (aHR: 0.74, 95% CI: 0.66-0.83). The reduced hazard of readmission did not significantly mediate (% mediated: 2.16%, 95% CI: À8.59%-11.93%) the association between receiving MHA service and all-cause mortality ( Figure S1).
Separating outpatient MHA services by physician type (primary care vs. psychiatrist) did not meaningfully change the association between receiving outpatient MHA care on the hazard of all-cause mortality (Table 3).
T A B L E 1 Descriptive statistics of cohort stratified by whether outpatient mental health and addiction (MHA) services were received in the 30 days following the index alcohol-related hospitalization. However, following stratification, there was no longer a significant association between receiving outpatient MHA services from a psychiatrist (aHR: 0.92, 95% CI: 0.83-1.01) or primary care physician (aHR: 0.99, 95% CI: 0.91-1.07) and hospital readmission, presumably due to (1) the small overall effect size of the association and (2) the reduced size of the exposure group following stratification ( Table 3). The reduced size of the exposure group likely also explains to the lack of significant association between receiving MHA services from both a psychiatrist and primary care physician and all-cause mortality (aHR: 0.77, 95% CI: 0.52-1.13), given that only 1.7% of the cohort received services from both physician types in the 30 days following discharge from the index alcohol-related hospitalization (Table 3).

| Characteristics associated with short-term readmission and mortality
Three thousand three hundred and fourteen (7.1%) of individuals who experienced an alcohol-related hospitalization in the study timeframe were excluded because they died or were readmitted to hospital in the first 30 days following discharge. Relative to those who were included in the cohort, these individuals were proportionately older (mean age: 56.98 years vs. 51.98 years, SD: 0.31), had a higher medical comorbidity burden (mean ADG score: 33.13 vs. 23.06, SD: 0.61), had more previous alcohol-related hospitalizations (% with 2+ previous alcohol-related hospitalizations: 28.0% vs. 8.2%, SD: 0.53), had a higher psychiatric comorbidity burden (% with prior inpatient psychiatric T A B L E 2 Receiving outpatient mental health and addiction (MHA) services in the 30 days following discharge from an alcohol-related hospitalization is associated with a reduced hazard of readmission and mortality-results from the multivariable time-to-event regression models.  Table 4). When the alcohol-related hospital readmissions and deaths from this subset of individuals were combined with those experienced by the main cohort, 19.1% and 11.5% of all individuals hospitalized for an alcohol-related health condition were readmitted to hospital and died (respectively) within 1-year of discharge in Ontario.

| DISCUSSION
Outcomes following alcohol-related hospitalizations in Ontario, Canada are poor, with one in five individuals being readmitted to hospital and more than 1 in 10 individuals dying within a year of discharge. Receiving physician-based outpatient MHA services in the first 30 days following discharge was associated with a lower hazard of alcohol-related hospital readmission and mortality; however, only one fifth of the cohort received these services. In turn, there is a gap in care that, if addressed, may be able to reduce the high rates of recurrent harm and death in this population. The association between follow-up MHA service use and all-cause mortality persisted regardless of whether the service was provided by a psychiatrist of primary care physician, indicating that a rapid referral to either physician type may help to reduce the risk of short-term mortality in this population. While this was the first study to characterize clinical outcomes following alcohol-related hospitalizations in Canada (apart from a previous study published by our group that made use of a comparable dataset to evaluate F I G U R E 1 Receiving outpatient mental health and addiction (MHA) services in the 30 days following alcohol-related hospitalizations is associated with a lower hazard of alcohol-related hospital readmission and all-cause mortality. (Left) Modeled cumulative incidence of alcohol-related hospital readmission in the year following the index alcohol-related hospitalization, stratified by whether or not an individual received outpatient MHA services in the 30 days following discharge. Derived from the unadjusted Fine & Gray subdistribution hazard model for alcohol-related readmission with death as a competing risk. (right) Modeled cumulative hazard of all-cause mortality in the year following the index alcohol-related hospitalization, stratified by whether or not an individual received outpatient MHA services in the 30 days following discharge. Derived from the unadjusted Cox proportional hazard model for all-cause mortality. By definition, an individual could not have died or been readmitted to hospital in the first 30 days following discharge, which is why the x-axes begin at 30 days.
T A B L E 3 The association between receiving outpatient MHA services and alcohol-related hospital readmission and mortality stratified by the type of service provider.

Service provider
n (% of cohort) rural-urban disparities in outcomes following alcoholrelated hospitalizations 11 ), the high rates of short-term readmission following alcohol-related hospitalizations align with the existing international literature. For example, Peeraphatdit et al. found a 23.9% 1-month readmission rate following hospitalizations for alcoholic hepatitis and Yedlapati et al. found a 58.8% 1-year readmission rate following hospitalizations for alcohol withdrawal. 4,9 Therefore, this study adds to a growing body of evidence that recurrent alcohol-related health service use is an important public health concern that needs to be addressed through improved individual-and population-level interventions. Indeed, the fact that individuals with 2+ previous alcohol-related hospitalizations were at a near four-fold higher hazard of 1-year readmission than those with none emphasized the existence of a "revolving door" of alcohol-related hospital admissions in Ontario, which has been described in other healthcare systems. 21 One strategy to reduce recurrent alcohol-related harm may be to increase the use of physician-based outpatient MHA services, as evidenced by the reduced hazard of alcohol-related hospital readmission and all-cause mortality among those who accessed these services in Ontario. Previous studies have come to similar T A B L E 4 Descriptive statistics of cohort stratified by whether alcohol-related hospital readmission or death occurred in the 30 days following discharge from the index alcohol-related hospitalization. conclusions but have been limited to specific clinical cohorts and specific types of outpatient addiction treatment. For example, Peeraphatdit et al. found that the use of alcohol rehabilitation programs within 30 days of discharge was associated with a $70% reduced odds of short-term hospital readmission and a $80% reduced hazard of mortality in a sample of $300 individuals hospitalized with alcoholic hepatitis. 4 Similarly, Gryczynski et al. found that case management, service linkage, and motivational support reduced the likelihood of readmission by an additional 25% beyond receiving addiction treatment alone in a sample of 400 hospital patients with an SUD (including but not limited to AUD). 22 The present study builds on this previous work by illustrating that early post-discharge use of any type of follow-up MHA care is associated with a reduced risk of readmission and mortality across all alcohol-related health conditions listed in the ICD-10. This suggests that there may be a role for a broader suite of mental health services (i.e., beyond just addiction treatment services) in reducing the risk of short-term harm following alcohol-related hospitalizations, which is not surprising given that $70% of the cohort had evidence of a comorbid psychiatric illness.
One aspect of increasing the use of follow-up MHA services will be to increase the physical availability of these services (allocating services to underserviced areas, training more mental healthcare workers, increasing clinical operating hours, etc.), as the current availability does not sufficiently meet the demand. 23,24 However, many individuals do not use MHA services even when they are physically available to them. Indeed, there are many upstream factors that contribute to development and progression of AUD (e.g., poverty, housing instability, alcohol access, hazardous drinking cultures) that relate to and interact with other systemic barriers to receiving MHA care (e.g., cost, stigma, biased referral practices, long wait-times, culturally inappropriate care). 25 Therefore, improving the use of MHA services following alcohol-related hospitalizations will require a two-pronged approach that both increases the availability of outpatient MHA services while simultaneously addressing the patient-and system-level barriers that are preventing people from accessing them.

| LIMITATIONS
We did not have access to data on private addiction treatment facilities or non-physician-based mental health services. This may have resulted in an underestimation of the number of individuals who received outpatient MHA services following alcohol-related hospitalization and biased the observed association between outpatient MHA service use and post-discharge readmission or mortality. However, individuals who received outpatient MHA care from a private source would have been placed in the "no outpatient MHA services" group, which, if anything, would have biased the observed association toward the null and resulted in an underestimation of the association between receiving outpatient MHA services and the study outcomes. Similarly, while the use of a landmark analysis was necessary to mitigate immortal time bias, it also meant that individuals who received MHA services after the 30-day cutoff would be categorized as not having received MHA care. This "late" care may also be associated with a reduced risk of readmission and mortality and, in turn, would have further biased the observed association toward the null.
The landmark analysis also limited the results of this study to only apply to those individuals who did not experience an adverse outcome in the first 30 days following discharge from the index event. This resulted in the exclusion of older individuals with more severe AUD (or related organ dysfunction) and a higher burden of medical comorbidities, and it is important to consider that the results of this study do not necessarily generalize to this subgroup of individuals. Follow-up studies on effective clinical strategies to reduce the risk of shortterm harm in this high-risk subpopulation are warranted.
Another potential limitation was the selection of a random event as the "index hospitalization" for those who had more than one hospitalization during the accrual window. This method deviates from a protocol that would be used in a prospective design, where individuals would generally be included at their first presentation to hospital. The rationale for this choice was to avoid a clustering of individuals with recurrent alcoholrelated hospitalizations at the beginning of the accrual window. Indeed, if the earliest hospitalization was chosen, this could create a temporal bias insofar as individuals experiencing a single alcohol-related hospitalization would be evenly collected across the full study timeframe whereas individuals with multiple alcohol-related hospitalizations (who foreseeably have different clinical and sociodemographic characteristics from those with a single hospitalization) would be preferentially collected earlier in the study timeframe. In contrast, the random selection method allows for all individuals to be collected evenly across the accrual window, which can be helpful when retrospectively analyzing alcohol-related health service use, as recurrent service use is common and could exacerbate this temporal bias. 11,26 Nonetheless, the deviation from a standard protocol of including individuals at first presentation to hospital (during the accrual window) should be considered when interpreting the results from this study.
Finally, treatment-seeking individuals who used MHA services had different demographic and clinical characteristics than those who do not, which may have contributed to the observed association between of receiving MHA services on the risk of short-term readmission and mortality. We controlled for many of these factors in our multivariable models; however, historical analyses of health administrative data come with an inherent risk of residual confounding. In turn, it is possible that there were unmeasured protective factors present among individuals who accessed follow-up MHA services that contributed to the associations observed in this study.
To conclude short-term outcomes following alcoholrelated hospitalizations in Ontario, Canada are poor with one in five individuals being readmitted to hospital and 1 in 10 individuals dying within 1 year of discharge. Receiving follow-up MHA care in the 30 days following an alcohol-related hospitalization was associated with a reduced hazard of 1-year readmission and mortality. However, only one fifth of individuals in Ontario received this care. In turn, improving rapid access to and use of outpatient MHA services following alcohol-related hospitalizations should be prioritized.