All‐cause and cause‐specific mortality in individuals with an alcohol‐related emergency or hospital inpatient presentation: A retrospective data linkage cohort study

Abstract Background and Aims Alcohol consumption is a leading risk factor for premature mortality globally, but there are limited studies of broader cohorts of people presenting with alcohol‐related problems outside of alcohol treatment services. We used linked health administrative data to estimate all‐cause and cause‐specific mortality among individuals who had an alcohol‐related hospital inpatient or emergency department presentation. Design Observational study using data from the Data linkage Alcohol Cohort Study (DACS), a state‐wide retrospective cohort of individuals with an alcohol‐related hospital inpatient or emergency department presentation. Setting Hospital inpatient or emergency department presentation in New South Wales, Australia, between 2005 and 2014. Participants Participants comprised 188 770 individuals aged 12 and above, 66% males, median age 39 years at index presentation. Measurements All‐cause mortality was estimated up to 2015 and cause‐specific mortality (by those attributable to alcohol and by specific cause of death groups) up to 2013 due to data availability. Age‐specific and age–sex‐specific crude mortality rates (CMRs) were estimated, and standardized mortality ratios (SMRs) were calculated using sex and age‐specific deaths rates from the NSW population. Findings There were 188 770 individuals in the cohort (1 079 249 person‐years of observation); 27 855 deaths were recorded (14.8% of the cohort), with a CMR of 25.8 [95% confidence interval (CI) = 25.5, 26.1] per 1000 person‐years and SMR of 6.2 (95% CI = 5.4, 7.2). Mortality in the cohort was consistently higher than the general population in all adult age groups and in both sexes. The greatest excess mortality was from mental and behavioural disorders due to alcohol use (SMR = 46.7, 95% CI = 41.4, 52.7), liver cirrhosis (SMR = 39.0, 95% CI = 35.5, 42.9), viral hepatitis (SMR = 29.4, 95% CI = 24.6, 35.2), pancreatic diseases (SMR = 23.8, 95% CI = 17.9, 31.5) and liver cancer (SMR = 18.3, 95% CI = 14.8, 22.5). There were distinct differences between the sexes in causes of excess mortality (all causes fully attributable to alcohol female versus male risk ratio = 2.5 (95% CI = 2.0, 3.1). Conclusions In New South Wales, Australia, people who came in contact with an emergency department or hospital for an alcohol‐related presentation between 2005 and 2014 were at higher risk of mortality than the general New South Wales population during the same period.

Measurements: All-cause mortality was estimated up to 2015 and cause-specific mortality (by those attributable to alcohol and by specific cause of death groups) up to 2013 due to data availability.Age-specific and age-sex-specific crude mortality rates (CMRs) were estimated, and standardized mortality ratios (SMRs) were calculated using sex and age-specific deaths rates from the NSW population.

INTRODUCTION
Alcohol is one of the most commonly used psychoactive substances globally.In 2016, 18.2% of people aged ≥ 15 years engaged in heavy episodic drinking (≥ 60 g alcohol in a drinking session) in the past month [1].Globally, alcohol was ranked the seventh leading risk factor for disease burden in 2016 [2].This pattern of consumption and harm associated with alcohol is reflected in Australia.One in four (25%) adult Australians drink alcohol at risky levels (> 40 g alcohol on a single occasion) at least monthly [3], and approximately 1.8% of hospitalizations and one in 10 emergency department presentations in Australia are estimated to be alcohol-related [4,5].
Alcohol consumption is one of the leading risk factors for mortality [6].People with an alcohol use disorder have a threefold higher risk of all-cause mortality compared to the general population [7,8].There have been studies on mortality risk associated with alcohol use in the general population [9] and in those receiving treatment for alcohol use disorder [7,8,[10][11][12].For example, in those seeking treatment for alcohol use disorder, there is a 10-fold increase in the risk of death from alcohol-related liver cirrhosis and mental disorders, and a high risk of death from unintentional injuries, suicide, cancers and cardiovascular diseases [12].However, less is known about mortality among individuals who may be experiencing problems related to their alcohol use who may not necessarily have engaged in formal treatment for an alcohol use disorder.This gap in knowledge is significant, given that only one-in five Australians with an alcohol use disorder receive treatment [13], with similar findings internationally [14].
Advances in data linkage of routinely collected administrative health-care data sets with mortality datasets can enable such study, facilitating powerful population-level analyses of mortality risk with fewer resource implications than cohorts established through primary data collection [15].Despite this, there have been few studies using such linked data to study mortality risk among people with alcoholrelated problems.Those that have been conducted typically focus on all-cause or specific causes of death (e.g.acute injury mortality [16][17][18]), or specific subpopulations of people presenting to health services with an alcohol-related diagnosis (e.g.people with a non-fatal injury with a comorbid alcohol use disorder diagnosis [19]).The exception is a data linkage study of a Danish population-based cohort of all adults with a first-time hospital contact for an alcohol-related diagnosis in 1998-2002 followed for mortality until 2012 [20].This cohort had four times the rate of all-cause mortality and 10-34 times the rate of alcohol-related mortality compared to the general population.This study demonstrates feasibility of using linked administrative data to estimate mortality risk in the broader population of people experiencing alcohol-related problems, highlighting capacity for similar study in other countries where alcohol is a public health priority, and scope to draw on other health-care data-sets (e.g. from emergency departments) where people with alcohol-related problems are heavily represented [4].
As such, the aims of the current study were to estimate the crude rates of all-cause and cause-specific mortality and excess mortality  [21].Access was approved by data custodians and linkage was undertaken by the Centre for Health Record Linkage (CHeReL) using Choicemaker software [22].Ethics approval was provided by the NSW Population and Health Services Research Ethics Committee [21].Alcohol-related diagnoses used for cohort identification are in Supporting information, Appendix S1.We excluded individuals identified solely on prenatal alcohol exposure or a diagnosis of alcohol in the blood without another alcohol-related diagnosis (Supporting information, Appendix S1).Individuals were also excluded if they had inconsistent information across data sets (e.g.date of birth/ death), were non-NSW residents or outside the age range 12-100 years at index presentation (see Supporting information, Appendix S2).

Mortality outcomes
All-cause mortality data on registered deaths (e.g.age at death, date of death) in NSW were extracted from the NSW Registry of Births, Deaths and Marriages (RBDM) from cohort commencement Mortality data were analysed as fully and/or partly attributable to alcohol (see Supporting information, Appendix S5 for codes); by the corresponding ICD-10 chapters (i.e.clustered based on body system and/or condition); and by subcategories derived from ICD-10 codes for mortality fully and/or partly attributable to alcohol (see Supporting information, Appendix S6 for codes).ICD-10 codes for mortality fully attributable to alcohol comprised alcohol-induced pseudo-Cushing's syndrome, Wernicke encephalopathy, mental and behavioural disorders due to alcohol use, degeneration of nervous system due to alcohol, alcoholic polyneuropathy, alcoholic myopathy, alcoholic cardiomyopathy, alcoholic gastritis, alcoholic liver disease, alcoholinduced pancreatitis and alcohol poisoning [24].ICD-10 codes for mortality partly attributable to alcohol were identified from the Australian National Alcohol Indicators Project and existing literature [20,[24][25][26].

General population mortality data
The COD URF data set for all deaths registered in NSW from

Data analysis
We allowed for 1 year of follow-up to capture out-of-hospital mortality (i.e.mortality that may not be directly related to the index presentation).To allow at least 1 year of follow-up period from cohort entry to death or censoring the two mortality analyses All-cause crude mortality rates (CMRs; per 1000 population) and indirect SMRs were calculated by age and sex using the observed and expected deaths to produce age-specific and age-sex-specific estimates.The expected deaths were calculated by multiplying the person-years accumulated in the study period by the age-sex-specific mortality rates (in the 12-15-and 16-19-year age groups, then in 5-year age groups from 15-19 to 85-89 years and 90+ years) of the standard population [27].A negative binomial model was fitted for number of observed deaths (in each sex and age group combination) as an outcome and number of expected deaths as exposure to estimate SMR and relative risk (RR) of sex (i.e.females compared with males).A Poisson model was fitted when there was no evidence of overdispersion (i.e.dispersion parameter was not significantly different from 0; see Supporting information, Appendix S7).The estimated SMRs and RRs from both the Poisson and negative binomial models are presented in Supporting information, Appendix S7.SMRs were computed as predicted values from the models.CMRs were calculated as the number of deaths in the cohort divided by person-years of observation.A Poisson model was used to estimate CMRs and RRs as the negative binomial distribution could not be used when there is only a single data point for computation of a crude rate.
Estimates based on numbers < 10 were suppressed in line with the data confidentiality protocol.
Similarly, cause-specific CMR, SMR and RR were estimated by sex, but not by age, due to small cell sizes after splitting by causes of death for some of the diagnosis groups.The regression modelling was performed in Stata version 16.0 (StataCorp).All other analyses were conducted in SAS version 9.4 [28].Findings are reported according to the REporting of studies Conducted using Observational Routinely-collected Data (RECORD) statement [29] (Appendix Supporting information, S9).Statistical significance is determined at P < 0.05.Analyses were not pre-registered; results should be considered exploratory.

All-cause mortality
Table 2 presents the excess mortality (SMRs), overall and by age and/or sex, and RR of excess mortality by sex.Mortality was consistently higher than the general population in all adult age groups and in both sexes (overall SMR = 6.2, 95% CI = 5.4, 7.2; Table 2).Overall SMRs were similar between the sexes (RR = 1.0, 95% CI = 0.7, 1.4).However, females had significantly higher SMRs in the 25-3, 35-4 and 55-64-year age groups and significantly lower SMR in the 85+ age group, compared with males in the corresponding age groups (Table 2).

Cause-specific mortality
Of the 154 552 individuals with cause-specific death data, there were 20 529 deaths throughout 626 770 person-years of observation.
Follow-up time ranged from 0 to 11 years, with a mean of 4.1 years (SD = 2.4).Table 3 presents the overall and sex-specific CMRs and RR of mortality of sex for specific causes of deaths in the cohort.CMRs by causes either fully or partly attributable to alcohol were 32.4 (95% CI = 31.8,32.9) in males and 17.4 (95% CI = 16.8, 18.0) in females.In general, females had lower RR than males for most of the causes.
T A B L E 1 All-cause mortality: crude (overall), and sex-, age-and age-sex-specific mortality rate (CMR; per 1000 population) and relative risk (RR) of sex for individuals with an alcohol-related hospital inpatient or emergency department presentation from 2005 to 2015 (n = 188 770).Relative risks significant at P < 0.05 are highlighted in bold type.Observed and expected number of deaths are available in Supporting information, Appendix S8.CI = confidence interval.
Total by causes defined as fully and partly attributable to alcohol  4).Crude mortality for causes fully attributable to alcohol was lower in females than males (CMR = 5.1 in females versus 10.0 in males; RR = 0.5, 95% CI = 0.5, 0.5; Table 3), but females had higher excess mortality from causes of deaths fully attributable to alcohol, compared with males (SMR = 110.2 in females versus 44.7 in males; RR = 2.5, 95% CI = 2.0, 3.1; Table 4).
Excess mortality was observed across all causes studied but was predominantly driven by diseases associated with chronic heavy alcohol use (e.g.alcohol-related mental and behavioural disorders, cardiovascular diseases, liver and pancreatic diseases).This accords with previous work [20], and reinforces the opportunity that acute healthcare services have to identify these individuals and intervene to address chronic patterns of alcohol use and accompanying health risks.Excess mortality was also observed for acute injury-related causes (e.g.suicide, interpersonal violence) arising from alcohol intoxication.This matches findings from a 1-year follow-up study of acute injury mortality among people in California who attended an emergency department (ED) and who had an alcohol use disorder (AUD) diagnosis, although SMRs were higher in the current study [16].
While small cell sizes prevented us from studying differences by age group, the aforementioned Californian study found very high risks for suicide and homicide in those aged 10-45 years [16].This suggests that strategies to reduce mortality risk may need to be tailored to age group.Similarly, the current study showed that women had greater excess acute injury mortality risk, while males had greater excess risk due to cardiovascular causes and certain cancers.This accords with existing literature [12,16,20] and reinforces importance of considering demographic differences in mortality risk, with scope in future work to consider other factors (e.g.socioeconomic deprivation [32]).

Implications
People presenting to emergency department and hospital with an alcohol-related problem are at increased risk of premature mortality.
This point of service contact may represent a unique opportunity to identify and intervene to potentially reduce such risk.Current Australian alcohol treatment guidelines recommend that emergency departments and hospitals should screen to identify risk alcohol use, offer brief interventions in certain circumstances and refer to general practitioner and other services for ongoing monitoring of alcohol use and interventions [33].
Given socio-demographic differences in excess mortality risk across injury and natural cause in cause-specific mortality risk, linking people with potential alcohol use disorder with person-centred, integrated care which addresses alcohol use and other co-occurring health issues (e.g.alcohol use alongside other mental health problems) may assist in reducing mortality risk [34].Understanding demographic differences in risk by sex can also help inform the shape of interventions; for example, these findings reinforce the importance of overdose and suicide prevention strategies for this population, particularly for females.In saying this, it is important to acknowledge the significant barriers facing health professionals in emergency departments and hospitals in screening, providing brief intervention and/or referring to other services for further treatment [35].
In-hospital alcohol consultation liaison services are established with the goal of improving identification of people experiencing alcohol-related harms and facilitating direct access to specialist services for support and treatment advice [36].Although resourcing of such systems is a challenge, alcohol consultation liaison services reduce the burden for front-line medical staff, and provide a clear pathway for ensuring appropriate care within hospital and postdischarge [36].Indeed, an economic analysis suggested that alcohol consultation liaison services in NSW, Australia, would save at least $100 000 per hospital per year on average by reducing the average length of hospital stay and decreasing frequency of admission over time [36].Similar outcomes have been observed for dedicated drug All causes partly attributable to alcohol 7.5 (6.1, 9.3) 7.6 (6.0, 9.8) 7.1 (4.9, 10.3) 0.9 (0.6, 1.5) Either fully or partly attributable to alcohol 8.0 (6.4,10.0) 8.1 (6.2, 10.5) 7.7 (5.2, 11.5) 1.0 (0.6, 1.5) By specific causes of death and alcohol brief intervention teams within emergency departments in other Australian jurisdictions [37], reinforcing the value of further study of outcomes from such services.Indeed, overall there is a need for research that evaluates the impact of various strategies deployed within these settings in reducing mortality risk to inform future policy and resource allocation.While our population-level sample was large, underascertainment of cases is probable, given that we could only determine alcohol involvement via diagnostic code and alcohol-related diagnoses are often not identified nor coded in emergency department and hospitals [38].This is particularly pertinent for emergency department data, where only one diagnosis field is available, so alcohol-related mortality risks may have been underestimated.Mortality rates may also be underestimated, as data on deaths that occurred outside NSW were not linked.There may also be a risk of bias from missing records due to people moving and/or dying in another jurisdiction.We attempt to mitigate this bias by computing SMRbased mortality data for all NSW residents.Finally, although we studied causes partly attributable to mortality, we cannot definitively attribute these to alcohol.

CONCLUSION
All-cause mortality among individuals with an alcohol-related hospital inpatient or emergency department presentation was substantially elevated over that of the general population.Cause-specific analyses showed that greatest excess mortality was in deaths caused by alcohol-related mental and behavioural disorders, liver cirrhosis, pancreatic diseases, viral hepatitis and liver cancer.Accurate estimation of excess mortality is important in informing health service delivery and policy planning in Australia.We need to evaluate and strengthen strategies to identify and support people with alcohol-related problems in hospital and emergency departments to reduce mortality risks.

(
including by age and sex where possible) in individuals who presented to hospital inpatient and/or emergency departments in New South Wales (NSW), Australia, with an alcohol-related diagnosis between 2005 and 2014.Given the varying mortality risk across age and sex, we estimated age-sex standardized mortality ratios (SMRs) to study excess mortality in the cohort in comparison with the general Australian population.METHOD Study design This is a retrospective cohort study of individuals identified between 2005 and 2014 and followed-up to 2015 to examine all-cause mortality, and individuals identified between 2005 and 2012 and followedup to 2013 to examine cause-specific mortality.Data are from the Data linkage Alcohol Cohort Study (DACS), a state-wide retrospective cohort study established through data linkage of persons in NSW, the most populous Australian state, with an alcohol-related hospital inpatient or emergency department presentation between 1 January 2005 and 31 December 2014.Details on cohort formation have been published elsewhere Individuals entered the cohort at the time of their first record ('index presentation') in the NSW Admitted Patient Data Collection (APDC) or Emergency Department Data Collection (EDDC) with an alcohol-related diagnosis between 1 January 2005 and 31 December 2014 (see Supporting information, Appendix S1).The index presentation is the first presentation within the given time-frame and does not necessarily constitute the person's first ever alcohol-related presentation.The NSW APDC comprised records of hospital inpatient separations in all public and private hospitals, public multi-purpose services and day procedure centres in NSW.A separation refers to a completed episode of care for an admitted patient, ending with discharge, death, transfer, leave against medical advice or a portion of a hospital stay beginning or ending in a change to another type of care.A principal diagnosis and up to 50 additional diagnoses are coded according to the International Classification of Diseases, 10th revision, Australian modification (ICD-10-AM).All diagnosis fields were analysed.The NSW EDDC comprises records of emergency department presentations in major metropolitan and non-metropolitan public hospitals in NSW.The number of emergency departments contributing has increased over time, noting that larger facilities take part, and so data are assumed to capture a substantial proportion of the NSW population [23].Only a principal diagnosis field is coded in the EDDC.This field was coded according to the International Classification of Diseases and Health Related Problems, 9th revision, clinical modification (ICD-9-CM), ICD-10-AM or the Systematized Nomenclature of Medicine-clinical terms Australian modification (SNOMED-CT-AU).

(i.e. 1
January 2005) to 31 December 2015.Cause-specific mortality data were obtained for these deaths from the Cause of Death Unit Record File (COD URF).Due to data processing lags, cause-specific mortality data were only available 1 January 2005 to 31 December 2013.The underlying and all contributory causes of death (up to 20 diagnoses) coded according to ICD-10 from the COD URF were analysed (Supporting information, Appendix S5).

1
January 2005 to 31 December 2015 by age (range = 12-100) and sex were obtained from the Australian Bureau of Statistics (2005) and the Australian Coordinating Registry COD URF (2006-15).
covered a different study period, as previously detailed.Personyears of follow-up were calculated from the index date of admission (cohort entry) until the date of death or censoring (all-cause mortality = 31 December 2015; cause-specific mortality = 31 December 2013).

From
the 208 143 individuals identified, 19 365 were excluded for eligibility reasons, leaving 188 778.A further eight with missing data on sex were excluded, therefore the all-cause mortality analysis included 188 770 individuals who entered the cohort between 1 January 2005 and 31 December 2014, with all-cause mortality data available to 31 December 2015.The cause-specific analysis excluded a further 34 218 individuals who did not have at least 1 year of follow-up data, resulting in the inclusion of 154 552 individuals who entered the cohort between 1 January 2005 and 31 December 2012, with causespecific mortality data available to 31 December 2013 (Figure 1).
for a cohort of individuals who presented with an alcohol-related hospital inpatient or emergency department presentation between 2005 and 2014 in NSW, Australia.All-cause mortality risk was six times higher than the general population overall.The greatest excess mortality risk arose from mental and behavioural disorders due to alcohol use, liver cirrhosis, viral hepatitis, pancreatic diseases and liver cancer.Females had twice the male excess mortality due to alcohol and other substance-related disorders, alcohol poisoning and suicides.Males had almost double the excess mortality risk for mortality due to liver cancer and hypertensive heart disease.This population experienced disproportionate mortality risk, often from causes which can be wholly or partially attributed to alcohol use; their attendance at acute health-care services represents a critical opportunity to identify their problem alcohol use and intervene to reduce mortality risks.This study was novel in its (i) use of multiple acute health-care service data-sets linked with mortality data, (ii) study of an Australian cohort and (iii) focus on a broader population of people with alcoholrelated problems than only those engaged in alcohol treatment.T A B L E 2 All-cause excess mortality: standardized mortality ratios (SMRs) and relative risk (RR) of sex for individuals with an alcohol-related hospital inpatient or emergency department presentation from 2005 to 2015 (n = 188 770).

T A B L E 3
Cause-specific mortality: crude (overall) and sex-specific mortality rate (CMR; per 1000 population) and relative risk (RR) of sex for the alcohol cohort from 2005 to 2013.
interval; -= not reported due to protocol of suppression of cell sizes n < 10.Relative risks significant at P < 0.05 are highlighted in bold type.aSee Supporting information, Appendix S5 for list of diseases counted towards causes attributable to alcohol; fully and partly are not mutually exclusive.T A B L E 4 Cause-specific excess mortality: standardized mortality ratios (SMRs) and relative risk (RR) of sex for the alcohol cohort from 2005 to 2013.

Limitations
There are several limitations to note.Data from the current study are drawn from one Australian jurisdiction (albeit the most populous one) from 2005 to 2015.Findings may not be generalizable to other jurisdictions or countries, nor to other time-periods.Future work using data over a longer and more recent time-period would assist in in identifying a person's first ever presentation and changes in clinical risk over time.In addition, future research could examine potential individual patient characteristic factors that may be associated with increased risks of mortality following an alcohol-related presentation.At presentation, health-care professionals collect a range of additional information about the patient in their medical records.Future research that makes use of this additional patient-level information collected to identify potential predictors of negative future outcomes are warranted.

Table 4
presents the excess mortality (SMRs), overall and by sex, and RR of excess mortality of sex, for specific causes of deaths.The cohort had elevated mortality in causes fully attributable to alcohol (SMR = 50.3,95% CI = 44.6,56.8;Table The SMRs were indirectly standardized by comparing to age-sex-specific deaths of the general NSW population.The estimated SMRs and RRs were computed in negative binomial models with number of observed deaths as an outcome and number of expected deaths as the exposure variable.RRs significant at P < 0.05 are highlighted in bold type.CI = confidence interval.