Main findings
This study was the first example of an analysis of the impact of ACTs on emergency hospital activity amongst an ARLD cohort at a national level using routine administrative datasets. We used the national PHE hospital alcohol team survey and English NHS hospital episode statistics to investigate the impact of ACTs on hospital activity for patients’ before and after diagnosis with ARLD. We proposed that presence of ACTs at hospitals where a first recorded diagnosis of ARLD was made would be associated with a reduction in emergency hospital activity when examining the difference in activity before and after this point in comparison to where there were no ACTs as a result of dedicated support to help these patients. However, using this approach where individuals in the ACT group may or may not have received the intervention, we did not find evidence of a change in emergency admissions (all cause or alcohol related) or all-cause emergency department attendances between hospitals with an ACT and those without a service.
Several other local evaluations of hospital alcohol services have shown ACT’s to be effective in reducing admissions, bed days, emergency department attendances and quality of care [
9‐
14]. These evaluations have focused on a single service and were able to restrict their evaluations to individuals who had definitely received an ACT intervention. Furthermore, they often used a local database or audit data to capture hospital activity for the patients followed up over time. The generalisability of the findings of these single service studies are limited as ACTs deliver care often designed and applied to the local context. Our current study, by contrast, wanted to assess the effect of ACTs at a national level in order to provide support for commissioning of preventative services.
Strengths and limitations
The cohort study design permitted the examination of ACT effects in a specific high need population. Our study was also more inclusive than previously published local evaluations in that it considered all secondary care activity within hospitals that allowed a comparative study to be undertaken at a national level and incorporated a robust statistical method to adjust for confounding factors.
It is important to consider this work in light of some limitations. We used the PHE survey of hospital ACT provision [
8] to determine which individuals in our ARLD cohort could have potentially received an intervention. We therefore did not know which individuals had received an ACT intervention. This limited the power of the study as it is reasonable to assume that not every individual presenting to an ACT hospital would have received an intervention because of limited capacity or working hours. Defining our incident ARLD cohort in 2009/10 was useful in that there was a substantial group of hospitals at this time which did and which did not have an alcohol team in place. We attributed responsibility for supporting the individual to the hospital where the patient had their first recorded diagnosis. We acknowledge that coding practice may vary between hospitals and therefore may not be an exact measurement of the point of diagnosis. This is expected to be a source of random error in our comparison groups as coding practice is unlikely to be influenced by the presence of an ACT. Once the hospital admission or outpatient appointment has ended, the individual may rely on referral and follow up by services in the community for on-going support. We had no information on community or primary care service provision in this dataset. In addition, where the index admission was away from a patients usual place of residence, we had no information to ascertain any follow up arrangements and where links are not in place between services it is likely that this may have been difficult. This was not accounted for in our analyses and may have been a factor in examining the effect of potential intervention by an ACT given our focus on emergency hospital activity.
From the available data, we were only able to make a dichotomous assessment of ACT provision. The PHE survey had collected some information on the models of alcohol services provided in hospitals, however, it was difficult to categorise these services into specific types due to substantial variation in the operational models and due to incomplete information. This included, for instance, the number of whole time equivalent staff in the team, whether a weekend and evening service was provided, whether the team was led by a senior clinician, whether the service targeted patients in emergency departments or specific wards. It was also possible that services may have evolved over time and the information in the survey reflected the snapshot in time when the survey was completed.
For pragmatic reasons, we used 2009/10 as the timepoint to define when a hospital had an ACT. The retrospective nature of the PHE survey we used to identify ACT provision does not allow us to accurately identify if an ACT was active in a particular hospital over the whole 2009/10 period.
There are a variety of drivers of emergency care activity, particularly in patients with ARLD. The provision or cessation of health and community support services within individual localities may have affected the results of this study, but our datasets have not allowed us to take these into account. However, any impact is likely to have been random across ACT and non-ACT groups due to multiple factors influencing local community provision of a service such as competing needs, priorities and availability of funding.
We have also not considered if the patients have developed more complex ARLD over time. There are limited clinical variables in the HES dataset and this has precluded us from measuring severity of ARLD through the Model for End-stage Liver Disease [
23] or Childs-Pugh [
24] scores.
Additionally, for those hospitals that we included in our category as not having a dedicated alcohol team, it is plausible that frontline staff may have delivered brief interventions as part of their routine practice. Brief interventions have been shown to be effective [
25] and so it is possible that this may have limited our ability to detect a difference between the hospitals which did and which did not have alcohol teams. It is also reasonable to assume that newly commissioned alcohol services would take some months to become fully effective as has previously been recognised [
9] and this was not accounted for in the analysis. There may have been other initiatives which we have not been able to consider in this analysis.
We used first recorded inpatient admission or outpatient appointment with an ARLD diagnosis as a proxy for the date of ARLD diagnosis. We also anticipated that diagnosis would be a trigger for offering support to the patient. Ideally, harmful alcohol consumption would be picked up before this point, however, even if this is the case, the individuals included in the cohort went on to receiving a diagnosis of ARLD and so may have benefitted from further support at this point. We recognise that hospital alcohol teams are unlikely to only focus on patients who have a diagnosis of ARLD and the patients that we included in this analysis were therefore at the severe end of the spectrum of alcohol related harm.
Explanation
There are several potential explanations for our findings. There may have been some important differences between the ACT hospital and non-ACT hospital cohorts. A larger proportion of individuals that had their first recorded ARLD diagnosis at an NHS hospital trust where an ACT was present were in the most deprived quintile of the English population and this may be an indication of greater need in this group. The higher all-cause emergency department attendances in ACT hospital trusts suggests these services were in areas with greater need. While our analyses accounted for time invariant confounders in the before and after measures (this would include deprivation over the two year period examined), it may have been that provision of an ACT avoided hospital activity that would otherwise have occurred in its absence. However, it is plausible that our measurement of exposure to an ACT was not suitably precise to detect the true effect of the intervention.
Interestingly, we observed an increase in average all-cause emergency department attendances per patient in the year following diagnosis regardless of whether the patient presented to an ACT or non-ACT hospital. We have already highlighted that improved recording of attendances in the A&E HES dataset over the time period we examined is a recognised limitation of these data (2,3). We also recognise there is local variation in coding practice which we would consider to be a source of random error in the measurement of A&E attendances at both the ‘year before’ and ‘year after’ time periods. However, it is also possible that this increase may reflect an increase in health seeking behaviour triggered by diagnosis of ARLD and lack of support or service provision in the community.
Future work
To overcome some of these challenges in examining the impact of ACTs within NHS settings on acute care activity, more precise measurement of exposure to ACT’s is needed. This could be achieved through working with a subset of hospitals to obtain details of ACT provision and use this information to restrict the analyses accordingly. Along with greater certainty of ACT provision, it may also be possible to gather data of individuals who received an ACT intervention and link these, maintaining anonymity of individuals, to HES data.
Furthermore, to overcome challenges of potential differences in baseline need in ACT and non-ACT groups, an alternative method, such as a retrospective matched control method, could be used to generate a matched control group from the same hospital population as the cases receiving an ACT intervention (where individuals who have received the ACT intervention are known). This method could be applied to replicate and potentially validate findings of a previous local evaluation demonstrating an effective service.
In order to replicate the methodology of this study with more precise measurement of exposure to an ACT, greater granularity is required in hospital data. For instance, a well-designed dataset, with appropriate linkage, collecting patient level information on when individuals are referred to a hospital alcohol team (or all substance misuse services), what intervention was received, the individuals alcohol consumption status and relevant diagnoses would facilitate this and ideally should be joined up with community service provision. Alongside this, repeated surveys by PHE to monitor the changing picture of alcohol teams in England’s hospitals and would enable more in-depth examination of which aspects of hospital alcohol teams optimise their effectiveness.
Whilst the Five Year Forward View (describing how the NHS in England needs to change to meet the current challenges and changing demands, published by NHS England in 2014) [
16] and the Government’s Alcohol Strategy 2012 (setting direction for alcohol policy in England and published under the Coalition government) [
26] identify the importance of preventative services, constrained public health resources will drive the need to demonstrate higher quality evidence of efficacy. The increasing health and economic burden of liver disease [
1] in the UK makes this a key priority for national preventative action.