Outcomes of first emergency admissions for alcohol-related liver disease in England over a 10-year period: retrospective observational cohort study using linked electronic databases

Objectives To examine time trends in patient characteristics, care processes and case fatality of first emergency admission for alcohol-related liver disease (ARLD) in England. Design National population-based, retrospective observational cohort study. Setting Clinical Practice Research Datalink population of England, 2008/2009 to 2017/2018. First emergency admissions were identified using the Liverpool ARLD algorithm. We applied survival analyses and binary logistic regression to study prognostic trends. Outcome measures Patient characteristics; ‘recent’ General Practitioner (GP) consultations and hospital admissions (preceding year); higher level care; deaths in-hospital (including certified cause) and within 365 days. Covariates were age, sex, deprivation status, coding pattern, ARLD stage, non-liver comorbidity, coding for ascites and varices. Results 17 575 first admissions (mean age: 53 years; 33% women; 32% from most deprived quintile). Almost half had codes suggesting advanced liver disease. In year before admission, only 47% of GP consulters had alcohol-related problems recorded; alcohol-specific diagnostic codes were absent in 24% of recent admission records. Overall, case fatality rate was 15% in-hospital and 34% at 1 year. Case-mix-adjusted odds of in-hospital death reduced by 6% per year (adjusted OR (aOR): 0.94; 95% CI: 0.93 to 0.96) and 4% per year at 365 days (aOR: 0.96; 95% CI: 0.95 to 0.97). Exploratory analyses suggested the possibility of regional inequalities in outcome. Conclusions Despite improving prognosis of first admissions, we found missed opportunities for earlier recognition and intervention in primary and secondary care. In 2017/2018, one in seven were still dying during index admission, rising to one-third within a year. Nationwide efforts are needed to promote earlier detection and intervention, and to minimise avoidable mortality after first emergency presentation. Regional variation requires further investigation.

planned/elective admission for ARLD would exclude a patient from your cohort or not.If not, there are presumably patients in your cohort who do have a previous diagnosis of ALRD, just not as an emergency.What is your justification for their inclusion in the cohort?P4 ll26-38.It would be helpful to include the full LAA definitions in the appendix as well as referring to the APT papers.P4 l59 please specify the codes (perhaps in appendix) used to define varices, ascites and comorbidities of interest.P5 l13 why only one year?P5 l25 Why was this limited to unplanned care?What is the justification for examining only a year's previous admissions?P6 l33 This is the only place you mention linkage to intensive care data.Please can you expand on this a little more -is this the Critical Care Minimum Dataset?P6 l4-8 Please specify (perhaps in appendix) a) which codes were used and b) whether only underlying cause of death or any cause of death was used.P7 ll41-44 'these figures support the assertion..' this is interpretive and belongs in discussion, not results.P7 ll50-52 "This leaves almost one in four with no preceding record of an alcohol-related diagnosis (despite subsequently being readmitted as an emergency with ARLD)" I have a few concerns with this statement: -The group with no preceding record of alcohol-related emergency admission in the last year are the 59% who have no emergency admission in the preceding year at all, plus the 24%x41%=~9% who had an emergency admission, so ~68% of the total.
-This isn't the same as saying they have no preceding record of an alcohol-related diagnosis: what about primary care and nonemergency admissions?What about diagnoses more than a year before the index admission?-In any case this kind of interpretive statement is again better in the discussion P7 ll58 onwards.Stage of liver diseasesee comments above.It would be supportive for the analyses that follow to move the K-M plot in supplementary figure 1 D to the main body.P8 ll14-21 This paragraph simply describes patterns in the figures without adding important information such as proportions and the results of statistical tests, which can only be found in the appendix table.Your statements "the proportion of first admissions with severe disease showed limited variation across deprivation quintiles" and "the proportion with "severe liver disease" (codes for cirrhosis or hepatic failure) was similar among men and women" seem to contradict the statistical testing reported in S1, which found a difference in the proportions significant at the p=0.001 level for both of these characteristics.Perhaps statistically significant but not large/important differences?P8 ll23-34.Perhaps with these small differences it would be useful to give age and percentages to one rather than zero decimal places in this paragraph and arguably could do so throughout paper.P8 ll34 P8 ll34-40 'Collectively these findings…' Interpretive and better in the discussion, not results.If changing coding practices over time is a genuine concern, would like to see more in-depth discussion of this and how it impacts on interpretation in the discussion section.P8 ll42 'Concerningly.' emotive and should be replaced.P8 l45.P=0.13 is really rather weak evidence for any change in the proportion of GP contacts with a record of alcohol, not sure it's valid to report it as a reduction.Separately, it would be useful to report the proportion of all patients who a GP contact with a record of alcohol/liver disease codes, i.e. with the denominator being all patients not just those with a GP contact in the last year.Otherwise, a decrease in the proportion with liver disease codes could for example be driven by an increase in proportion with a GP contact in last year, but no decrease in the overall recognition of liver disease P9 l57 How did you deal with patients transferred between hospitals when calculating length of stay?P9 l17 when you refer to deaths within 48 hours of admission, does this mean within 1 day of admission or 2 days of admission?Note that a death within 2 days of admission does not necessarily mean within 48 hours, since neither time of admission nor death is available in these data, e.g.patient admitted 3am Monday and died 3pm Wednesday.Use of 'just 48 hours' emotive.P9 ll45-47 Or died before transfer to ICU? Suggest move this to discussion section anyway.P10 l5.I can't find a Table 3 in the main ms.P10 ll14-16 Again think an additional decimal place would be useful on the CFRs P11 ll5-10.Please put results of sensitivity analyses in appendix Discussionsee comments above.

Brown, Cristal
The University of Texas at Austin, Internal Medicine REVIEW RETURNED 27-Jul-2023

GENERAL COMMENTS
Peer Review -Outcomes of first emergency admissions for alcoholrelated liver disease in England over a ten year period: Retrospective observational cohort study using linked electronic databases These authors acknowledge the gap in the literature regarding characterization of this population on a nationwide level iii.
They utilize an algorithm that they have previously tested and manually confirmed accuracy to improve the ability to detect patients with ARLD and link to available hospital administrative and death data.iv.
Study aims are cleared defined.c.Methods i.
The research design is appropriate to address the study aims ii.
Although here is sufficient description of the data sources, case definitions, and outcomes, I would recommend clarification of the following items: 1.Under "Population of interest", provide an explanation for defining index admission as having no such admission within the preceding 10 years as opposed to no prior admissions at all. 2. Under "Patient-level covariates derived from the index admission", why was hepatic encephalopathy not included as a co-variate for advanced liver disease?Ascites, variceal hemorrhage and HE are the 3 major complications associated with decompensated liver disease and the omission of this symptom should be addressed.3.Under "Patient-level covariates derived from the index admission", it would also be helpful to explain the decision to separate CCI categories of 0-1 and ≥ 2. Is this standard or provide an explanation for this choice iii.
There are no concerns with the statistical methods utilized and their methodological description."Amongst these is that inherent to the use of CPRD: as vividly demonstrated in the Atlas of Variation of Liver Disease (uncited in this report) there is major national geographic variation in liver disease incidence and outcome and these important regional effects are unexplored."HES-linked CPRD was the only dataset suited to the aims of the present study since we wished to examine primary care contacts during the year before index emergency hospital admission to explore potentially "missed opportunities" in the care pathway.HES-linked CPRD is the only source of nationally-representative linked data between primary care electronic records and HES.Our work focused on reporting at "national" aggregate level for the CPRD population and has important implications for interpreting routine national statistics for advanced liver disease in the UK.We believe it is essential that methods for cohort discovery and characterisation of ARLD are improved and that "getting it right" at national-level is a prerequisite before drilling down to local levels.National trends are of most relevance to an international readership and we note Reviewer 2 did not raise this particular point.We regard the full national HES dataset as better-suited to exploring local or institutional-level variation but it cannot be linked to primary care events.The "Atlas of Variation of Liver Disease" uses the full national HES dataset.It contains various metrics related to "cirrhosis" of any aetiology (not ARLD per se) and uses standard approaches to case definition.Based on our past work and data presented in the current paper, that data source requires careful interpretation.Such statistics do not take full account of the contribution of non-primary coding patterns or unspecified stages of liver disease to true burden and mortality from "cirrhosis".Nevertheless, we agree that there's abundant evidence of geographical variation in the incidence of alcohol-related harms based on population-level statistics and of a strong association between prevalence with socioeconomic factors.Our present work is more focused on care processes and care outcomes (i.e.what happens to patients in the year leading up to admission and what is the outcome of acute hospitalization?).We show that our primary metric of interest (case-mix adjusted CFRs for first admissions) is not associated with area-based deprivation status per se but rather with patient-level markers of "severity".This justifies our inference that observed national trends in this measure may reflect improvements in hospital care at this key milestone in the care pathway, albiet modest and without a major national shift in underlying trends post-NCEPOD etc. 2 However, to further address this comment we have added some focused exploratory regional analyses.These examine in-hospital case-fatality rate.The methods, results and discussion sections have been revised to include these new data.Word limits mean that most of this data has to appear in Supplementary Materials.
In our revised discussion, we further emphasize the limitations of CPRD and the need for further research.The present report already provides a large amount of novel national-level data.A more extensive exploration of regional variations is beyond the scope of a single paper.Further research is ongoing and we hope our work stimulates others to seek to improve routine statistics, such as the 'Atlas of Variation', by adopting algorithm-based methods."A further and similarly important limitation is the assignment of liver disease severity based upon primary codesaccurate assessment of liver disease severity is a challenge for practicing hepatologists, and extrapolation from HES codes even more uncertainparticularly since more than a quarter of cases had an unspecified stage of ARLD.The authors mention both these limitations in the discussion, but I felt these warranted greater emphasis."Firstly, we would emphasize that the assignment of severity was not determined from "primary codes".This statement implies that staging was based exclusively on whichever ICD-Code had been recorded as the "primary" diagnosis (i.e. the code appearing in the first diagnosis field of HES).This was not the case.Our algorithmic method screened all coding positions to find the relevant liver-related code and takes the "recorded-stage" from this.The method for assigning recorded-stage takes advantage of the LAA, whereby the liver-related code can appear in a lower diagnostic position (provided the primary diagnosis is relevant to an emergency admission for ARLD).Hence, a patient with a primary diagnosis of "Jaundice" followed by a secondary diagosis of "K703 Alcohlic cirrhosis of the liver" would be flagged as We report the frequent recording of 'unspecified' (K70.9)code rather than a stage-specific code in HES in "first" admissions.This is entirely predictable from a clinical perspective (especially at "first" admission) but has significant implications for interpreting the literature and current public health metrics.Typically, cases coded with K70.9 are missing from cohort definitions of "advanced liver disease" or "cirrhosis" based on administrative data, such as the the Atlas of Variation or published research.Within this often-excluded group, we show a significant prevelance of markers of advanced liver disease (ascites and varices).We demonstrate their contribution to overall case volume and mortality burden and present data on their survival relative to sub-populations with a stage-specific discharge code.These data are an essential component of the present paper, highlighting the limitations of making simplistic assumptions about disease stage and the potential merits of algorithmic approaches for cohort discovery and phenotyping.For the final risk-adjustment models of case fatality the reviewers will note that we adopt a simplfied binary variable for recorded-stage (Liver Failure vs All Other Codes) to mitigate this phenomenon and we use additional "flags" of liver disease severity derived from other codes (Ascites and Varices).Author reply: We focused specifically on GP contacts and unplanned admissions within the preceding year for two reasons.Primarily, this was because such encounters were judged to be close enough to the index admission to assume that all patients could be assumed to have an established history of excessive alcohol consumption (if only they had been screened effectively).Given the natural history of ARLD, we proposed that the absence of any relevant code in these contacts "within the year" of index emergency admission would imply missed detection of harmful alcohol-use.This blanket assumption of "missed opportunities" becomes less credible for healthcare contacts that occured in the more distant past.Secondly, with respect to GP contacts, a one-year period of prior CPRD practice registration is the minimum period for eligibility.Hence, not all CPRD patients have more than one-year of continuous general practice data recording before their "index" date.Variability in the duration of continuous CPRD practice registration is a recognised limitation of this dataset.We did not examine more distant GP records, which might cover varying periods of follow-up.   .Your statements "the proportion of first admissions with severe disease showed limited variation across deprivation quintiles" and "the proportion with "severe liver disease" (codes for cirrhosis or hepatic failure) was similar among men and women" seem to contradict the statistical testing reported in S1, which found a difference in the proportions significant at the p=0.001 level for both of these characteristics.Perhaps statistically significant but not large/important differences?"Author reply: We had tried not to place too much emphasis on quite small (but statistically significant) differences.The relevant analyses and stats are in Supplementary Materials and we do not believe these deserve greater emphasis given the volume of other data presented."P8 ll23-34.Perhaps with these small differences it would be useful to give age and percentages to one rather than zero decimal places in this paragraph and arguably could do so throughout paper."Author reply: Done 7 "P8 ll34 P8 ll34-40 'Collectively these findings…' Interpretive and better in the discussion, not results.If changing coding practices over time is a genuine concern, would like to see more in-depth discussion of this and how it impacts on interpretation in the discussion section".Author reply: We have expanded the relevant part of the discussion."Coding drift" is an ever-present possibility in all studies using HES data over long periods but most authors simply ignore or fail to report time-trends in covariates that might be vulnerable to evolving coding practice.The published literature using HES is replete with papers that apply the CCI score as a risk-adjuster without considering time-dependent variation in depth or completeness of discharge coding.Our sensitivity analyses excluding comorbidity entirely did not change the key conclusions in relation to time trends.The same was the case for models excluding recorded-stage (Liver Failure vs Not).See our revised commentary and expanded Supplementary Materials.P8 ll42 'Concerningly.' Emotive and should be replaced.Author reply: Revised."P8 l45.P=0.13 is really rather weak evidence for any change in the proportion of GP contacts with a record of alcohol, not sure it's valid to report it as a reduction.Separately, it would be useful to report the proportion of all patients who a GP contact with a record of alcohol/liver disease codes, i.e. with the denominator being all patients not just those with a GP contact in the last year.Otherwise, a decrease in the proportion with liver disease codes could for example be driven by an increase in proportion with a GP contact in last year, but no decrease in the overall recognition of liver disease" Author reply: Re-phased and clarified.The lack of significant change is taken to demonstrate a lack of "improvement" over time (i.e.no significant increase).The focus here is on "missed opportunities" during primary care contacts within a year of index admission.We assume a history of excessive alcohol consumption would well-established at this time and elective diagnosis of underlying liver disease possible.Please see the extensive list of READ codes we used, which sought evidence of any hint of alcohol or liver problems being suspected during those prior contacts.As with the data on prior emergency admissions, the relevant denominator is the proportion of GP consultersi.e.what proportion of those who were seen at the practice during the year before admission had any alcohol, or liver, codes recorded in those encounters."P9 l57 How did you deal with patients transferred between hospitals when calculating length of stay?" Author reply: We did not explicitly look at inter-hospital transfers.Length of stay is not a major focus of the paper.We don't believe that accounting for a minority of such transfers would have a significant impact on aggregated national-level trends.

GENERAL COMMENTS
We thank the authors for their comprehensive response to the review.Overall, our concerns have been satisfactorily addressed.A few minor comments remain which may be worth further consideration.
It may be useful to make it explicit in the "Population of interest: First emergency admissions for ARLD" paragraph that an elective admission in the preceding ten years would not exclude inclusion of the patient.There are a number of statements in the ms which could still imply it is the first admission overall and might usefully be revised.
The opening sentence of the discussion 'Our study represents the first detailed description of characteristics and outcomes of index emergency admissions in the English population' might better be expressed as ; 'Our study represents the first detailed description of characteristics and outcomes of index emergency admissions with ARLD in the English population'

Brown, Cristal
The University of Texas at Austin, Internal Medicine REVIEW RETURNED 27-Oct-2023

GENERAL COMMENTS
The authors have provided clear justifications and acknowledgement for each point of feedback.They have adequately addressed all items and updated the manuscript as appropriate.I have no concerns with moving forward with publication.
The "Data Sources" paragraph has an error in April 1997 to Mark (March) 2021 that should be updated.Methods -Case definitions for ARLD admissions -change alterntive to alternative

VERSION 2 -AUTHOR RESPONSE
Reviewer: 1 3 "I also felt that a greater emphasis should be placed upon a key message of the report -' making every contact count' though active screening for problem alcohol use.There unacceptable proportion of cases presenting for the first time with advanced stage disease even when there have been prior contacts in primary and / or secondary care with other medical issues.As part of this discussion, it would be helpful for the more information on the natural history of ARLD.How long would a patient need to have been using alcohol harmfully to develop ARLD at the various stages you describe?How long are the earlier stages of ARLD and what symptoms might be being missed by GP/previous hospital visits?"Reply: The are a number of key messages, including the positive findings of improving prognosis of admissions over time.However, we have revised the manuscript accordingly.Our conclusions in the abstract and final part of discussion place strong emphasis on earlier recognition and intervention, citing relevant literature for readers.We have expanded the methods section to emphasize the rational for focusing on events within a year of emergency admission for ARLD."Other Comments: P4 ll18-19.Please can you clarify whether a previous planned/elective admission for ARLD would exclude a patient from your cohort or not.If not, there are presumably patients in your cohort who do have a previous diagnosis of ALRD, just not as an emergency.What is your justification for their inclusion in the cohort?"Reply: Our study focuses on first emergency admission for ARLD (i.e.non-elective).We were interested in studying the case mix and outcomes of patients hospitalized acutely for a first time with this condition.We have proposed previously that a person's index unplanned hospitalization for ARLD represents a critical point in the care pathway and a pragmatic focus for developing better metrics of initial acute care derived from hospital administrative data.In the paper's title, and throughout the manuscript, we emphasize that our primary focus was on emergency admission.This key milestone in the patient journey may often represent the point of first-ever contact with liver services and/or the point of first-ever formal diagnosis of ARLD.However, our report focuses on the processes and outcomes of the index emergency admission for ARLD, as opposed to seeking to identify a subcohort of first emergency admissions which were a verifiable point of "first-ever" diagnosis.Identifying index emergency admissions does not specifically require screening of elective admissions.However, this comment is relevant to one of the sentences in our original discussion (paragraph 3), where we had stated that ARLD is 'frequently diagnosed for the first time during unplanned admission'.We have revised and better-qualified this statement.There are only a limited number of reasons why a patient with established ARLD might be admitted "electively" to hospital for planned treatment of liver disease per see.g. either for daycase endoscopy (for management of varices) or daycase paracentesis (for treatmentresistant ascites).In our experience, such elective admissions are almost invariably downstream of one or more previous emergency presentations.Nevertheless, we accept that a patient's first emergency admission for ARLD might have been preceded by a non-elective admission in some cases.This does not alter the main focus of our paper (i.e.characteristics and outcomes of index emergency admissions).We would expect prior discharge diagnoses of ARLD to be known and reflected in the primary care record.We report that only 14.7% of consulters in primary care during the year prior to admission had READ codes for liver disease (i.e. a prior elective diagnosis).We believe this would include patients with well-known liver disease who had received prior elective hospital 4 care (and yet escaped any prior emergency hospitalisation).However, we acknowledge the perceived limitation in our revision."P4 ll26-38.It would be helpful to include the full LAA definitions in the appendix as well as referring to the APT papers."Reply: The LAA code lists are published and freely-available "open access".The relevant link to download the code lists has been added to the manuscript.To encourage appropriate citation of our original work, we prefer to sign-post that existing link rather than replicating publication in another supplementary file.https://onlinelibrary.wiley.com/action/downloadSupplement?doi=10.haveadded the citation to give appropriate attribution.We have updated the methods and references.To focus on non-liver comorbidity, we simply excluded "mild" liver disease and "moderate or severe" liver disease categories.• Charlson ME, Pompei P, Ales KL, MacKenzie CR.A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.J Chronic Dis.1987;40(5):373-83.doi: 10.1016/0021-9681(87)90171-8.PMID: 3558716.These extensively utilised code lists are given below (for reviewer's convenience): Myocardial infarction: I21.x, I22.x, . The expanded Methods section futher emphasizes our focus is on "missed opportunities" among patients who had a contact in the year before admission.The denonimator population for these figures are the people who were admitted as an emergency.Our purpose was to illustrate that when prior emergency hospitalisations occurred within just a year of admission for ARLD, an alcohol-related diagnosis was "missing" in 1 in 4 cases."P7 ll58 onwards.Stage of liver diseasesee comments above.It would be supportive for the analyses that follow to move the K-M plot in supplementary figure1D to the main body."Author reply: Yes, agreed, this deserved more emphasis.See our earler comments relating to the importance of reporting data by "recorded-stage".We have moved the relevant KM curve accordingly, incorporating it into Figure 2 in the revision (Fig 2H)."P8 ll14-21 This paragraph simply describes patterns in the figures without adding important information such as proportions and the results of statistical tests, which can only be found in the appendix table with the conclusions and limitations stated by the authors.