Determinants of care outcomes for patients who die in hospital in Ireland: a retrospective study

Questionnaires 1 & 2 were completed within a week of the patient’s death. Questionnaire 3 was completed three months after the death, similar to the minimum bereavement period adopted in other surveys of bereaved relatives [22-24]. Questionnaire 4 was completed by a sample of 10 nurses and healthcare assistants in each of the wards where a patient’s death was included in the study. Questionnaire 5 was completed by a quota sample of 100 staff in each hospital with participation proportionate to five different staff categories, excluding nurses and health care assistants. Questionnaire 6 was based on 2008 data and authorised by hospital management before being returned.

Multilevel modelling (MLM) was used to analyse the influence of care inputs on each care outcome since, in addition to separating individual-level (L1), ward-level (L2), and hospital-level (L3) influences, MLM also controls for covariance between the care inputs. A total of 13 multilevel models were generated by analysing the five care outcomes and the different perspectives of nurses, doctors and relatives. The model for each care outcome also contained an estimate of how that outcome was related to the other care outcomes. All models were estimated using the software package MLwiN v.2.10, and specified as three-level linear regression equations. The significance level (p < .05) was set in order to maximise the chances of identifying statistically significant influences on care outcomes. In order to build a more complete picture of these influences, ANOVA was used to test for differences between the means of each care input.

The sample of 999 deaths was mainly drawn from acute hospitals (880, 88%) with the remainder (119, 12%) from community hospitals. As a proportion of total deaths in Ireland in 2008, this represents 10% of acute hospital deaths and 29% of community hospital deaths, making it the largest study ever undertaken in Ireland to assess the quality of care provided to patients who die in hospital.

Data was collected from respondents (999 nurses, 737 doctors, 461 relatives, 2358 ward staff, 1858 hospital staff) on care inputs and care outcomes (Table 1). The response rate by nurses and doctors was 84% and 68% respectively (based on those hospitals which could have met the full quota of 50 deaths) which yielded a sub-sample of 737 deaths with matching data from both nurses and doctors. The response rate by relatives was 46% and within the range found in similar surveys of relatives, both in Ireland [25,26] and elsewhere [22]; this yielded a sub-sample of 461 deaths with matching data from nurses, doctors and relatives. The response rates by hospital staff to Questionnaire 4 (83%) and Questionnaire 5 (64%) were relatively high.

There is a statistically significant relationship between each of the care outcomes (Table 2). This suggests that these outcomes may represent different aspects of the same underlying concept. The largest correlations are between patient care, acceptability of the way patient died and family support.

Differences in how nurses, doctors and relatives assess care outcomes were measured by the level of agreement between their assessments of individual patients (Table 3). The results show that the highest level of agreement between nurses, doctors and relatives was for acceptability of the way the patient died (68%); for all the other care outcomes, the level of agreement between nurses, doctors and relatives was no greater than 25% (Table 3). There was a higher level of agreement between nurses and doctors but, for all but one care outcome, the level of agreement between them was no greater than 45%. Differences were also evident in the assessment of pain. The proportion of patients deemed to have pain ‘all or most of the time’ during the last week of life varied between doctors (11%), nurses (17%) and relatives (34%) (Table 4).

Most patients, based on the complete sample and the three sets of assessments, are reported to be relatively comfortable as far as pain (84-90%), nausea (94-95%), anxiety (87-89%), restlessness (83-85%) and chest secretions (80-83%) are concerned, but a smaller percentage are able to breathe comfortably (60-65%). Patient care, when expressed using the original 10-point scale, was 8.1 (according to doctors), 7.5 (according to nurses) and 7.3 (according to relatives). The proportion of ‘unacceptable’ deaths, scoring 3 or less out of 10, was higher in the assessment of relatives (21%) than nurses (13%) or doctors (3%). Family support, expressed using the original 10-point scale, was rated 8.3 according to nurses and doctors, and 7.0 according to relatives.

Most variation in care outcomes is explained by L1 variables with relatively little influence exercised by L2 and L3 variables (Additional file 2); this is partly due to the small sample sizes at L2 (283 wards) and L3 (43 hospitals) relative to the requirements of multi-level modelling. Eight sets of care inputs have a statistically significant influence on at least one care outcome: disease and sudden death; route of admission; physical environment; team meetings; communication; facilitating relatives; staff readiness; hospital governance.

Cancer patients have better care outcomes than other patients because they are more likely to be planned admissions (though most are still unplanned), to die in a single room, to be the subject of more team meetings and better communication, to have relatives who stayed overnight and were present at the moment of death. These are statistically significant influences on care outcomes and, because directly observed by nurses, doctors and relatives, are likely to have practical clinical significance. By contrast, the quality of end-of-life care received by dementia patients is poorer because they are more likely to die in a multi-occupancy room where there is less and poorer communication and where relatives are less likely to be present at the moment of death. The study suggests that if all patients were offered the same standard of care that is currently available to cancer patients, then the quality of end-of-life care in Irish hospitals could be improved significantly.

Our analysis suggests that the system of care for patients at the end of life is essentially emergency-led rather than electively-led and, within that, there are systematic variations which make certain wards (oncology and geriatric) and medical specialties (cancer) more conducive to a planned approach to the needs of patients. When these findings are added to the acknowledged under-provision of community-based and home-based services for medical needs in Ireland, and the fact that EDs are the only route of admission to hospital for most patients, it becomes clear that route of admission is symptomatic of more general problems of accessing services for patients at the end of life, with consequent knock-on impacts on overall care outcomes. Based on this study, a more planned route of admission, which would also require better coordination of services between hospital and community, could significantly improve care outcomes at the end of life.

A significant finding is that the quality of nurses’ discussion with relatives, rather than patients, was more influential in the nurses’ assessment of care outcomes. This may be due to the way communication was measured which focused on verbal communication since that may be less important than non-verbal communication as patients in the last week begin to show signs of withdrawing from the world and from contact with those around them. At the same time, previous research suggests other possible explanations for the patterns of communication identified: (i) there may be a tendency among health professionals to speak with the families of older people rather than the older person [30] (ii) hospital practitioners may have difficulty talking about dying and death [31,13] (iii) there may be a fear that relatives have a power to complain which dying patients do not [32]. Whatever the reason, these findings suggest the need for some deeper reflection by healthcare professionals on whether the patterns of communication revealed by the study are entirely consistent with a patient-centred approach and the importance of respecting the patient’s autonomy as far as possible.

The study shows that experience and training combine to improve the readiness of nurses to care for dying patients. The effects of readiness on care outcomes are visible not only to nurses but to doctors and relatives as well. This finding draws attention to the importance of training but also of retaining experienced nurses within the hospital and ward, and ensuring they have a direct role in patient care where they can have an impact on patient outcomes and on their colleagues.

The study shows substantial differences in how nurses, doctors and relatives assess pain with correspondingly low levels of agreement between them. This suggest the assessment of pain may lack specificity (resulting in false positives) and/or sensitivity (resulting in false negatives) although it is impossible, from this data alone, to estimate the number of true positives or true negatives. The prevalence of pain (‘all or most of the time’) is lower than reported in a previous study in Ireland [25] and lower than studies of elderly patients in long-term care in Europe [33], the US [34] and Canada [35] where about 50% of the patients experienced pain in the last week of life; in roughly half of these cases, the experience of pain was a daily occurrence. This evidence, in conjunction with the diverging views of relatives, nurses and doctors on the prevalence of pain, gives some grounds for questioning how well pain is diagnosed and treated among patients who die in Irish hospitals.

The main strength of the study is that it is based on a large sample of hospital deaths and covers many aspects of the patient’s final journey in hospital, ‘from admission to discharge’. An additional strength is that it is based on the judgements of nurses, doctors and relatives who were with the patient during the last week of life. A significant weakness is that the patient’s voice is missing. Moreover the finding that nurses, doctors and relatives are often at variance in their assessment of care outcomes implies that their judgements may not necessarily coincide with those of patients.

It is noteworthy that hospital staff tend to give consistently higher ratings for all care outcomes compared to relatives which suggests that care outcomes may not be as good as nurses and doctors believe. It is possible that nurses and doctors assess care outcomes not just by reference to explicit standards but also use implicit standards which may be self-referential, based on limited information and may risk positive bias. There may also be a ‘study effect’ whereby hospital staff provided overly-positive ratings for care outcomes due to a sense of pride in their work, a fear of negative consequences or out of organisational loyalty.

Finally, since this is a cross-sectional study it is important to state that identification of statistically-significant effects between independent variables (care inputs) and dependent variables (care outcomes) does not necessarily imply a causal relationship.