In keeping with previous work on risk factors for readmission this study shows that age, lung function, and previous hospital admission are the most powerful predictors of readmission for AECOPD. However, this is the first prospective study to report that depressive symptoms recorded after discharge are associated with increased risk of readmission over 365 days and that an individual marker of socio-economic deprivation (owner-occupiers) are associated with the frequency of hospital readmission for AECOPD in patients cared for by nurse led EDS. Additionally, we found that compared with those not readmitted, readmitted patients had poorer HRQOL and worse depression and anxiety at 12 months follow-up.
Whilst hospital readmission might be unavoidable and necessary for some patients with AECOPD, psychosocial factors might account for a significant proportion of potentially unnecessary readmissions among the most vulnerable patients managed by EDS. Unlike Ng et al  we did not find an association between mortality and depression, mainly because our study was not adequately powered to examine this endpoint. However, in accounting for our findings about readmission possible mechanisms might relate to the finding that frequent exacerbators are more depressed than infrequent exacerbators , and depressed patients have a higher risk of exacerbation and possibly hospital admission . The precise relationship between psychological distress and exacerbation risk has yet to be adequately addressed however. The majority of studies that have tested associations between psychological distress and exacerbation risk are methodologically weak, leading to inconsistent and heterogeneous results. The best available evidence suggests that psychological distress might confer greater risk for symptom based as opposed to event-based exacerbations that demand therapeutic interventions and/or possibly hospital admission, but this conclusion warrants further scrutiny from well designed prospective studies . Furthermore, risk factors for relapse associated with treatment failure of a first exacerbation may well be different to risk factors associated with readmissions following successful treatment of exacerbations that are discrete events.
It is surprising that we did not find a similar association between anxiety and readmission. Panic attacks and panic disorder are the most common anxiety disorders among COPD patients owing to heightened physical arousal following catastrophic negative cognitions of ambiguous physical sensations such as shortness of breath . However the HADS does not specifically measure panic disorder and item 7 related to panic demonstrates item bias for severity of illness among COPD patients . Because we used the HADS we therefore might have been unable to detect whether panic attacks were associated with risk of readmission.
Social factors are also important determinants of healthcare utilisation. There is growing evidence from ecological studies that socio-economic status is a key driver of respiratory hospital admissions, but evidence from individual level measures of socio-economic deprivation is more equivocal . Contrary to expectation we did not find an association between the Carstairs deprivation index and risk of readmission, despite the fact that the Carstairs index is known to outperform individual social class as a measure of deprivation . This may have been due to the absence of variation among Carstairs scores among this study population, suggesting that this measure is prone to the ecological fallacy.
We did however show that patients who owned and occupied their homes had significantly fewer readmissions. Home ownership, especially in the UK, is a well established indicator of material living standards and long-term cumulative wealth . Our findings are in-keeping with other studies that have shown that home ownership exhibits stronger associations with health than conventional markers of socio-economic status such as income, education, and occupational and social class . Further, housing-related health hazards (e.g. damp, cold) typically associated with poorer quality of housing among renters is associated with poorer health , and housing conditions have been implicated with respiratory health status and hospital admission .
Unlike socio-economic status, low perceived social support was not associated with risk of readmission. This sample reported high levels of social support at baseline and ESSI scores remained stable throughout follow-up. Over the life course of a chronic illness such as COPD, perceived social support fluctuates - it grows or decays in the presence or absence of acute stressors . This sample was regularly exposed to stressful events i.e. exacerbations and admissions, and these events may have contributed to the maintenance of high levels of social support at 3 and 12 months. In addition, the relatively high scores on the ESSI may have reflected the fact that only patients with adequate social support or were self-caring were eligible for referral to EDS at the study sites. An alternative methodological explanation points to the fact that whilst the ESSI only contains generic questions related to perceived social support, this instrument might have a different factorial structure when used with COPD patients. The ESSI might therefore measure either different constructs or measure perceived social support differently in COPD patients.
By contrast, psychological health and HRQOL deteriorated at 12 months for patients who were readmitted, but remained stable among patients not readmitted. The reasons for this trend are not clear because we were unable to assess whether changes in psychological distress (anxiety and depressive symptoms) and HRQOL led to increased risk of readmissions, or whether increased readmissions led to a change in psychological distress and HRQOL. Among COPD patients, psychological health is known to deteriorate over time and these changes are only weakly correlated with changes in physiological parameters, highlighting the fact that COPD exerts its effects in body systems other than the lungs and requires multi-dimensional assessment .
This study included a self-selected sample drawn from UK EDS and patients who declined to participate may well have had more elevated symptoms of anxiety and depression than those who did enrol in the study. This group may also have represented an atypical group of patients who were willing to take part in a research study and their responses may have been affected by social desirability bias. Additionally, we did not have ethical approval to collect data from non-participants, limiting opportunities to compare characteristics such as age, sex and smoking status between refusals and participants. For these reasons the results may not be generalisable to wider COPD populations and settings where EDS is not available. However, whilst patients on EDS represent only a well supported sub-set of the total COPD patient population, they are among the most vulnerable and use a disproportionate amount of healthcare resources . We therefore specifically recruited patients on EDS to identify psychosocial characteristics that might be associated with risk of readmission among a group of vulnerable patients that are high intensity users of healthcare.
Despite targeting patients on EDS in three hospitals this study had low uptake and the results should therefore be deemed exploratory and interpreted cautiously. The reasons for under recruiting are uncertain. Patients who receive home based interventions such as EDS might perceive the prospect of additional home visits from researchers as too intrusive and burdensome, but there is little evidence for this. Indeed, on the contrary, patients with chronic illness, especially those with severe or end stage illness, may well welcome opportunities to engage in research as it might confer therapeutic benefits  and appeal to a sense of altruism . These perceived benefits may well have motivated patients to participate in this study. The more plausible reason for under recruitment stems from the fact that the research team were not sufficiently embedded within the clinical teams, leading to over reliance on specialist nurses, who were not ordinarily research active or afforded research time to identify and recruit eligible patients. When working with non-research active clinicians, recruitment needs to be incentivised, either financially, or by securing continuing professional development credits for time spent on research activities.
The principal investigator collected data at all time points during this study and was not blind to patients' baseline psychological status, potentially leading to measurement bias. However, patients' psychological health status was measured using self-report rather than investigator led questionnaires limiting opportunities for the principal investigator to introduce measurement bias. Additionally, all follow-up questionnaires were scored by an independent research administrator at the end of the 12 month follow-up period. In this way the principal investigator was not exposed to patients' follow-up scores on the HADS, SGRQ or ESSI, potentially minimising measurement bias during the study follow-up.
We conducted three different separate regression analyses: logistic regression to assess which predictors were associated with readmission within 365 days, Cox regression to assess which predictors were associated with time to first event (readmission or death), and Poisson regression to assess which predictors were associated with the number of readmissions. The final regression models included a set of a priori predictors (age and sex) and measured the effects of as many known confounders within sample size limits. We avoided overfitting the models and preserved degrees of freedom, an approach which is transparent and potentially replicable in other samples . The models demonstrated that worse lung function (FEV1) was associated with all three outcomes, depression score was associated with readmission within 365 days, and home ownership was associated with fewer admissions. A larger sample size may have allowed us to explore these apparent relationships further, and perhaps with more power, we would have arrived at more consistent results over the three regression models.
We did not enter into the models data on treatments, including medications for psychological distress which would have allowed us to examine whether treatment for anxiety and/or depression reduced the risk of readmission. Further, we did not collect data on whether patients had received pulmonary rehabilitation which is known to improve anxiety and depressive symptoms in some COPD patients . Resource constraints did not allow for longitudinal assessment of lung function (FEV1), thus limiting opportunities to examine whether changes in disease severity were associated with deteriorations in patient reported outcomes. However, in recognition that COPD is a multi-component disease with systematic manifestations, we prospectively measured patient reported outcomes such as HRQOL which are possibly more important markers of disease progression and severity than lung function alone . In addition we also collected both individual and area-based measures of socio-economic status which is a methodological strategy known to capture important differences in social and economic status . Collecting individual level data on home ownership is particularly relevant to research involving patients who are typically older and/or retired adults for whom the importance of other indicators of socio-economic status is much reduced .
Identifying and improving the health of patients with complex long term conditions at most risk of unplanned and unnecessary secondary care has become a key objective of advanced healthcare systems such as the NHS . Case finding tools now exist to identify patients at high risk of hospital readmission within 12 months , but isolating specific characteristics associated with unnecessary or preventable readmissions has proven difficult. Identifying deteriorations in psychological distress may improve current case finding models for patients at risk of hospital readmission.
Designing interventions to reduce the risk of hospital readmission for AECOPD remains a challenge however. As with EDS, there is little evidence that nurse-led chronic disease management programmes that offer variations of case-management reduce readmissions . Nor do written action plans with minimal or no self-management education . In designing interventions that enhance the management of AECOPD patients out of hospital there is scope for evaluating interventions that target identifiable and modifiable risk factors . Psychosocial factors such as depression are potential therapeutic targets and nurse-led minimal psychological interventions have proven effective in managing depression in chronically ill elderly patients . In the UK, the National Institute for Health and Clinical Excellence have published guidelines that incorporate stepped care models to facilitate the delivery of accessible and effective treatments for treating depression in people with chronic physical illness . However, despite evidence that supports the efficacy of antidepressants and structured forms of psychotherapy, depression remains under-detected and under treated, especially among older adults and in patients with medical comorbidity . There is thus scope to design and evaluate models of care that might improve the detection and management of psychological distress among vulnerable COPD patients discharged home and reduce demand for unnecessary and unscheduled hospital admissions.