Population and data source
The National Survey of Bereaved People (VOICES, Views of Informal Carers - Evaluation of Services) is a nationally representative cross-sectional survey which was conducted in England annually for 5 years, 2011–2015, to collect information about the quality of end-of-life care, particularly in the last 3 months of life [30]. The survey was commissioned by NHS England and administered by the Office for National Statistics (ONS). The survey’s results are based on a relative’s or friends’ perspective on the quality of end-of-life care provided to the decedent. Previous analyses of VOICES data have been used to inform national policy on end-of-life care service and assess and evaluate the quality of end-of-life care in different settings (home, hospital, care homes and hospices) [31]. Every year, a stratified sample of around 49,000 adults was selected from deaths which are registered in England. The VOICES questionnaire was sent by post to the person who registered the death of the decedent who is normally a family member or a close friend. Respondents were contacted once between 4 and 11 months after death (two further reminder questionnaires were sent if there was no response). The sampling weight and non-response weight were created by the ONS for each year. The sampling weight and non-response weight were then combined by taking the product of the two. We used the combined weight to adjust for sampling and non-response biases. Further information on VOICES methodology is available from the ONS [30].
Sampling
We obtained data from five annual VOICES surveys conducted between 2011 and 2015. During this period, 246,763 people were invited to participate, of whom 110,311 (45%) returned a completed questionnaire. Because VOICES survey does not contain information about palliative care in settings other than home, we examined factors associated with good pain relief at home in the last 3 months of life for people with advanced progressive illness. In addition, the main outcome was the success of pain relief at home which was applied only to decedents who had pain at home in the last 3 months of life.
The following exclusion criteria were applied:
-
1.
Decedent who died suddenly or were not ill prior to death.
-
2.
Decedent who did not spend any time at home in the last 3 months of life.
-
3.
Decedent who did not have any pain at home in the last 3 months of life.
Independent variables
Respondent characteristics included age, gender and relation to the decedent (spouse/partner, son/daughter, other). Decedent characteristics included age, gender, cause of death (cancer or non-cancer), index of multiple deprivation (IMD) quintiles (1 = most deprived, 5 = least deprived), duration of illness before decedent died.
We developed variables for service characteristics labelled receiving specialist palliative care at home in the last 3 months of life (yes or no), recorded preference for place of death (yes or no) and urgent care provided out of hours (once or twice, three times or more, not at all).
Specialist palliative care proxy measure
Respondents were asked about the decedent, ‘when he/she was home in the last three months of life, did he/she get any help from any of the services: hospice home care nurse or specialist, hospice at home service, Macmillan nurse or Marie Curie nurse? (Macmillan and Marie Curie are UK-based charities that fund clinical nurse specialists in palliative care who deal with cancer and non-cancer patients). If respondents answered ‘yes’ to any of the three questions, it was assumed that their relative had received specialist palliative care in the last 3 months of their life, defined as professionals or services whose core activity is providing palliative care. If responders answered ‘no’ to all three questions, it was considered for the purposes of this research that their relative had not received specialist palliative care in the last 3 months of their life.
Recorded preference for place of death
The survey asks about the decedent ‘Did she/he ever say where she/he would like to die?’ If responders answered ‘yes’, they were asked about the preferred place of death. After that, they were asked ‘Did the health care staff have a record of this?’ Having a preference recorded for place of death in the medical records was used as a positive indicator for this variable.
Outcomes
The primary outcome was the extent of pain relief at home. We collapsed the response categories included in the survey for ease of interpretation into: (1) good pain relief (pain relieved completely, all of the time and completely, some of the time) and (2) poor pain relief (pain relieved partially, not at all).
Statistical analysis
Based on individual-level survey data, numbers and percentages (both unweighted and weighted) were calculated to summarise decedent and respondent characteristics. Logistic regression models were used to examine the association between decedent and respondent characteristics and good pain relief. All variables that had a p value less than 0.1 univariately (to account for potential collinearity) were included in an initial multivariable model. As the aim of the analysis was to identify factors associated with good pain relief, rather than develop the most parsimonious model, variables were retained in the final multivariable model if they improved the fit of the model based on the likelihood-ratio test (using backward selection p < 0.05). In the final multivariable logistic regression model, we assessed potential multicollinearity using the variance inflation factor (VIF). Statistical software IBM SPSS statistics version 24 was used for data management and analysis. We used the published weights for the VOICES survey in the analysis to account for the study design and to adjust for non-response bias [30].