Recruitment of participants within hospice at home and service models
A total of 339 dyads were recruited from 12 HAH services. The target sample size of 66 patients was reached in each service model; there were four HAH services in models 1 and 2, three HAH services in model 4 and one in model 3 (). The number and proportion of dyads by model ranged from 75 (22.1%) in model 4 to 103 (30.4%) in model 1; the number of dyads from individual hospices ranged from three to 81 (see ). Over-recruitment of patient–carer dyads (compared with the original protocol) was undertaken because it became clear as the study progressed that the QODD completion rate was lower than had been predicted; this amendment received ethics approval.
Breakdown of participant numbers by service model and hospice
The IPOS: mean total scores by service model at recruitment
Data availability for analysis
Although 339 dyads were recruited, data were not available for all participants for all outcomes. Of the 339 patients recruited, 284 (83.8%) died during the study period. A DOD was established (either reported by carers post bereavement or obtained from the HAH service) so that time in the study from recruitment to death could be calculated.
A post-bereavement interview to collect responses to the QODD was given by 132 carers (46.5%) of the 284 people who died during the study period.
Of the 339 dyads recruited, 221 (65.2%) provided service use data by completing one or more AHCRs. Of these 221, it was possible to include 178 (80.5%) in the analysis of costs because they died during the study period and had a known DOD.
The availability of data on key outcome measures is summarised in . There were 327 dyads for which different combinations of outcomes were available. Twelve dyads provided no AHCRs (service use information) and no QODD because the patient was still alive at the end of the study. There were 43 dyads that had supplied at least one AHCR but for which the DOD was not known, and there were 67 dyads for which the DOD was known but there were no AHCRs or QODD data.
Number of dyads providing different combinations of outcome and service use data
Characteristics of participants at recruitment, by service model
There were no differences in background sociodemographic characteristics of participants recruited across the four models, except that carers in models 1 (larger, 24/7 services) and 4 (smaller, not 24/7 services) were more likely to be partners (rather than friends) of patients, and they were significantly older than the carers in the other models (see Appendix 9).
Clinical measures collected by hospice staff at recruitment showed differences between the models, with patients in models 3 (larger, not 24/7 services) and 4 (smaller, not 24/7 services) having worse health status than those in models 1 (larger, 24/7 services) and 2 (smaller, 24/7 services). Scores for the IPOS for each service model are shown in and , confirming the significantly worse health status of participants in models 3 and 4 at the point when they joined the study. The severity of reported outcomes was also explored. Of the 337 non-empty IPOS forms, 306 (90.8%) had at least one ‘severe’ (score 3) or ‘overwhelming’ (score 4) response [model 1: 88/102 (86.3%); model 2: 70/79 (88.6%); model 3: 79/81 (97.5%); and model 4: 69/75 (92.0%); chi-squared test, p = 0.058]. Individual IPOS items by service model are given in Appendix 10, Tables 31–48.
The IPOS score at recruitment: mean with 95% CIs for each service model. CI, confidence interval; Q, quadrant.
Data at recruitment on functional status (Karnofsky score: 0% = dead . . . 30% = almost completely bedfast . . . 100% = normal) identified patients in models 2, 3 and 4 as being sicker at baseline. The proportions of patients in the ‘deteriorating’ and ‘dying’ categories of the patient Phase of Illness measure were also highest in models 2, 3 and 4 ().
Patient functional status and Phase of Illness scores by service model at recruitment
There were also differences between models in the time (days) between recruitment to the study and death. Of the 284 patients who died during the study period, those in model 1 were in the study for significantly more days (mean 87 days) than those in the other three service models (overall mean 56.2 days), but there was considerable variability. The 36 patients still alive at the end of the study had been recruited, on average, 267.8 days beforehand ().
Patient time (days) in study by service model
Outcomes
Primary outcome: the Quality of Dying and Death questionnaire
A QODD was returned by the carers of 132 of the 284 patients (46.5%) who died during the study. Four respondents answered fewer than six of the 30 questions from which the total score is derived; they were excluded from the analysis. Response rates by service model and by individual HAH service are shown in Appendix 11. The proportion of carers who returned a QODD was significantly higher in model 1 than in the other service models. The frequency with which each of the 30 QODD questions were answered is shown in Appendix 12. The baseline sociodemographic and health status variables of the 152 carers who did not provide a QODD differed from those of the 132 who did only with respect to the education levels of the carer and the patient, which were higher among those responding than among those not responding (data not shown).
The total mean QODD score [range 0 (terrible) to 100 (almost perfect)] was 66.25 (SD 21.98, median 70.74, range 0–100, IQR 54.1–82.0) (n = 128). The distribution of the total QODD score () is not bell-shaped, with a slightly extended left-hand tail and a right-censored right-hand tail, the latter being caused by 100 being the maximum possible score. The distribution does not preclude the use of parametric techniques to analyse the total QODD score, but renders the median a slightly more appropriate summary statistic to describe the distribution alone. A score of 70 is associated with ‘a good death’.100
Histogram of total QODD score (with superimposed normal distribution).
Secondary outcomes
Patient achieved preferred place of death
Of the 284 dyads for which the patient died during the study period, information on both the PPOD and the actual place of death was available for 222 (78.2%); of these, 162 (73.0%) (shaded in ) achieved their PPOD and 60 (27.0%) did not. If care/nursing home is interpreted as the patient’s home (although it was not when the PPOD was collected at recruitment), then the percentage who achieved PPOD rises to 75.7% (168/222) (see ).
Preferred place of death by actual place of death
For the 62 patients who died without having expressed a PPOD, the places of death were distributed as follows: home, n = 29 (46.8%); hospice, n = 17 (27.4%); hospital, n = 8 (12.9%); care/nursing home, n = 4 (6.5%); and unknown, n = 4 (6.5%). Comparing the 58 patients who had a known place of death but an unknown PPOD with the 222 patients with a known place of death and a known PPOD, there was no statistically significant difference in the percentages dying in hospital: 8 out of 58 (13.8%) of the former versus 20 out of 222 (9.0%) of the latter (chi-squared test: p = 0.280).
Post-bereavement carer satisfaction: Views of Informal Carers – Evaluation of Services 1 and 2
Of the 284 eligible dyads, 132 (46.5%) carers returned a completed QODD, of whom 127 (96.2%) completed VOICES 1 and 128 (97.0%) completed VOICES 2. Responses are shown in . The median response to VOICES 1 was 1 (the best) (i.e. agreement that the carer and family got as much help and support from the health and social care services as they needed); for VOICES 2, it was 2 (i.e. that the carer rated the help and support they had received as excellent). Treating the 1–5 scales as linear, the mean for VOICES 1 was 1.31 (SD 0.64) and the mean for VOICES 2 was 1.98 (SD 1.06).
Responses to VOICES 1 and 2, post-bereavement service satisfaction
Service satisfaction and carer burden in the last 28 days of life
Completion of each AHCR included a question on satisfaction with services (scaled as follows: 1, exceeded expectations; 2, just met expectations; and 3, fell short of expectations) and a question on carer burden score (range 0 = best to 24 = worst). The last responses prior to the death of the patient (provided it was recorded within 28 days of death) were analysed (i.e. 143 service satisfaction ratings and 149 carer burden scores). The mean satisfaction and carer burden scores were 1.52 (SD 0.66) and 7.78 (SD 6.12), respectively. Histograms are shown in and .
Histograms of service satisfaction (with superimposed normal curve). a, Responses for last 28 days of life, for which 1 = exceeded expectations, 2 = just met expectations and 3 = fell short of expectations. (more...)
Histograms of carer burden (with superimposed normal curve). a, Responses for last 28 days of life; mean of six questions for which 0 = best and 24 = worst.
Comparison of outcomes by service model
Outcomes are summarised by service model in . Outcomes by individual HAH service are shown in Appendix 13. There was a statistically significant difference in the QODD scores between the four service models. Comparison of both mean scores and using the dichotomised scoring [≤ 70 (terrible, poor, intermediate) vs. > 70 (good, almost perfect)] reveals that carers in model 3 reported a significantly worse death than those in models 2 and 4 ( and ). Responses to each item of the QODD by individual HAH service are shown in Appendices 13 and 14 (including Tables 52–81). Significant differences were found between service models on service satisfaction (worst in model 1) and carer burden in last 28 days (least in model 1); model 2 scored marginally better than the other models on VOICES 1 and 2. There was no statistically significant difference between models in achievement of PPOD, although the proportion was higher in model 2.
Summary of outcomes by service model
Total QODD score by service model
Costs per service model: all formal care. Boxes show IQR with median bar inside. Whiskers are minimum to maximum, after excluding all outliers. Outliers (small circles) are 1.5–3 IQRs from the end of the box; extreme outliers (asterisks) are > 3 (more...)
Total QODD score using dichotomised scoring by service model
Factors associated with the Quality of Dying and Death questionnaire score (primary outcome)
Bivariate tests were used to explore the association between the QODD scores (available for 128 dyads) and a range of potential influencing factors (listed in Appendix 15): patient and carer characteristics; place of death and achieving PPOD; length of time the patient was involved with the hospice; three items from the QODD reflecting the condition of the patient, communication with the carer during the last 7 days and how prepared the patient was for death; and the number of days between death and completion of the QODD (which could affect carer adjustment to life after the loss of a loved one).
Results revealed that higher total QODD scores (i.e. a better quality of dying and death) were associated with:
A lower total QODD score was associated with the patient being in the study for a longer time (days) (Pearson’s correlation r = –0.217; p = 0.014). No other statistically significant associations with the total QODD score were found.
Exploratory regression analysis to investigate the effect of service model on outcomes
Seven outcome variables were modelled using the variables listed in Appendix 16. Four of these outcomes were gathered from the post-bereavement interview [total QODD score, achieved a good death (i.e. QODD score of > 70), VOICES 1 and 2]; two were gathered from the last AHCR completed prior to death, provided it was within 28 days of death (service satisfaction in the last 28 days, carer burden in the last 28 days); and achieving PPOD was gathered from the baseline assessment at recruitment and HAH service records of death.
Associations between the total QODD score (primary outcome) and secondary outcomes indicated positive correlations with VOICES 1 and 2 and service satisfaction in the last 28 days, but not with carer burden in the last 28 days (). The total QODD scores were available for 77 (47.5%) of the 162 patients who achieved their PPOD and for 25 (41.7%) of the 60 who did not achieve their PPOD. There was no difference in the mean QODD scores between these groups: 70.1 (SD 18.5) in the PPOD achieved group versus 61.2 (SD 20.0) in the PPOD not achieved group; unpaired t-test: p = 0.162. Reducing the QODD score to its dichotomous form, a good/almost perfect death (score of > 70) was reported for 46 out of 77 (59.7%) who achieved their PPOD, compared with 11 out of 25 (44.4%) who did not (chi-squared test: p = 0.168).
Associations between primary outcome (QODD total score) and secondary outcomes
Linear or logistic regression modelling was performed, with service model entered initially (model 3 was withheld as baseline) and additional predictors then determined by forward stepwise selection. Results in the form of fitted parameters for service model and for all other statistically significant predictors are summarised in , with interpretations provided in Box 1. Results of the logistic regressions are presented as adjusted odds ratios.
Regression modelling of outcomes with summary of fitted parameters from final models
Final model interpretation
Modelling of the two QODD outcomes produced the most robust findings and the highest levels of explanatory power. Both revealed that improved quality of death was associated with small providers (models 2 and 4), dying at home or in a hospice and when the patient was aware for a longer time that they were dying. The QODD total score model also showed that carers who were female and carers who were educated to university level were more likely to report a good death.
Achieving PPOD was less likely when there was a longer time between recruitment to the study and death. This may be because the PPOD was recorded at recruitment and patients can change their minds. This would be consistent with the finding that patients in model 2 were both (on average) in the study for less time and more likely to achieve their PPOD.
Regarding views on help and support received, carers of service model 2 reported more help and support in the post-bereavement interview and that it was of better quality (VOICES 1 and 2). Service model 1 was associated with less satisfaction in the last 28 days. Lower quality of help and support (VOICES 2) and levels of satisfaction in the last 28 days were reported by carers whose patients died in hospital. The carers of female patients reported lower levels of support (VOICES 1) and university-educated carers provided better ratings (VOICES 2).
Carer burden was less in the last 28 days when the patient died at home, and in service model 1 [in which patients (on average) had longer associations with the hospice].
Service-level analysis of resources and costs
Information on the resources involved in running each hospice and costs was sought through interviews with case study site managers during phase 1. Most hospices found it difficult to provide the detailed data on human and other resources that would be needed for a costing analysis. Either the information was not easily available to them or sharing it posed confidentiality issues. In the case of variables such as the ratio of nurses to HCAs, respondents were sometimes unsure of staff grades; they also reported that staffing levels and composition changed frequently, as professionals joined and left the service during the study. Based on interview responses, the home care delivered by each HAH service was designated as led by either RNs or HCAs. Except for the three smaller HAH services in model 1 (Acacia, Camellia and Echinacea) and one HAH service in model 4 (Marigold), all sites were led by HCAs (see Appendix 9).
Patient-level analysis of service use
Data on service use for the patient-level analysis were gathered through the AHCR, which was completed by interview at baseline (with reference to the previous 2 weeks). Telephone interviews (mostly with carers) to collect subsequent service use data were intended to take place every 2 weeks, up to the death of the patient. In some cases, carers were unavailable or busy and interviews could not be undertaken to schedule. When patients were in the study for many weeks or months, administering AHCRs became burdensome and the interval between interviews was extended.
The AHCR requested information on service contacts inside the home (i.e. community nurses/DNs/HCAs, hospice nurses/HCAs, home/personal care workers, GPs, allied health professionals, social workers, representatives of voluntary organisations) and outside the home (outpatient appointments at hospitals/hospices/clinics, visits to A&E, Day care), inpatient stays (hospital, hospice and care/nursing home), telephone calls, medications, supplies and equipment, and informal caring. At the end of the AHCR, there are three items for recording carer burden (six items, each on a five-point scale), the patient’s functional status (ECOG Performance Status Scale, six levels: 0 if fully active, 5 if dead) and satisfaction with services (whether they had exceeded, met or fallen short of expectations) over the period covered by the data collection.91
Completion of Ambulatory and Home Care Records
Of the 339 patient–carer dyads in the study, 221 (65.2%) provided at least one AHCR, of whom 178 had a known DOD and could be included in the economic analysis (see ). The distribution of the 178 patient–carer dyads was not even across service models: most (n = 68, 66.0%) were in service model 1 and the fewest were in service model 2 (n = 27, 33.8%). The progress of patients through the study and the provision of AHCRs (service use data) by service model and by individual HAH service are shown in Appendix 17, Tables 83 and 84.
In total, 613 AHCRs were returned for the 178 patients who had died and provided at least one AHCR (mean 3.44 per patient, maximum 23). Almost half of the total AHCRs available for analysis were in service model 1 (304/613, 49.6%); only 78 out of 613 (12.7%) related to model 2. As health and social care use changes in the months, weeks and days prior to death, the service use data were broken down initially into seven periods of number of days before death (0–7, 8–14, 15–21, 22–28, 29–92, 93–182 and ≥ 183 days). The dates covered by AHCRs did not coincide with these time periods, so an allocation algorithm was designed and implemented (explained in detail in Appendix 2). The number of AHCRs available by hospice and time period is shown in Appendix 18. The number of patients with AHCR data by service model and time period is shown in Appendix 19. Because an algorithm allocated AHCR data to study time periods, the number of AHCRs and patients in any time period may not be the same; this is explained in the footnote to Table 86 in Appendix 19. The pattern of AHCR completion reflected recruitment to the study, with most AHCRs relating to service model 1, which had recruited the most participants and recruited them furthest from the time of death (see ).
As the availability of AHCR data diminished considerably after the period 29–92 days, subsequent analysis of service use and costs was conducted for three time periods: the last 2 weeks before death (days 0–14), the penultimate 2 weeks before death (days 15–28) and between 4 weeks and 3 months from death. Beyond 92 days from death, the number of AHCRs returned was < 10 in three of the four models; therefore, it was decided that analysis in this time period was not viable.
Service use
shows summaries of service use in each of the three time periods: ‘nursing’ (RNs and HCAs) and social care home visits, all formal care contacts (health, social and voluntary sectors, in and out of the home, except for inpatient care), total inpatient nights (hospital, hospice, care/nursing home) and informal care. Informal care was measured in hours per day (rather than visits/contacts). (Full details of service use, broken down into individual items by time period and model, are shown in Appendix 19.) Data are presented as mean, SD, median and maximum number of visits per day (computed as the number of visits in the time period divided by the number of days covered by the AHCR in the period). The number of patient–carer dyads reporting zero contacts is also shown for each item.
Summary of main items of service use, by time period
The largest item of formal service use for all service models and for all time periods was nursing and social care delivered in the home. Three sources of nursing and other care support were distinguished in the data (community nurses/DNs and HCAs from local NHS services, nursing and HCAs from HAH services, and care workers provided by social services). Because there is substitutability between these three groups, and respondents were not always sure which source a visiting professional had come from, the data are presented with the three groups combined, to provide a measure of the total in-home nursing and social/personal care support received by patient–carer dyads (see ), and separately (see Appendix 20). Very little in-home contact with voluntary services was reported in any site (see Appendix 20).
Across the whole sample, there was a general trend for rates of formal and informal care to be higher in the periods closer to death (see ). The median number of hours of informal care in the last 2 weeks of life was > 20 hours per day (total of main carer and any additional carers). The median number of nursing and personal/social care visits was 1.76 per day; the median number of total health, social and voluntary sector contacts, including nursing but excluding inpatient stays, was 2.36 per day. On average, the number of inpatient nights (hospital, hospice and care home) was low. There was a lot of variability between individuals in all categories of service use (see ).
On average, respondents reported using two items of equipment (e.g. beds, bedding, hoists, commodes, walking frames, shower seats). Many also reported use of continence pads and food supplements (data not shown).
Differences between service models
The main summary measures of service use [in-home nursing/carer visits, total formal care contacts (including nurses, excluding inpatient care), informal caring hours] by model are shown in box plots in –. Box plots for individual HAH services are shown in Appendix 21, –.
Service use by model: total community health and social care visits per day. This includes DNs/community nurses, and HAH staff and personal carers, and includes night sits. Boxes show IQR with median bar inside. Whiskers are minimum to maximum, after (more...)
Service use by model: informal caring. Boxes show IQR with median bar inside. Whiskers are minimum to maximum, after excluding all outliers. Outliers (small circles) are 1.5–3 IQRs from the end of the box; extreme outliers (asterisks) are > 3 (more...)
Service use by model: total visits excluding inpatient stays and informal care. Boxes show IQR with median bar inside. Whiskers are minimum to maximum, after excluding all outliers. Outliers (small circles) are 1.5–3 IQRs from the end of the box; (more...)
Inspection of the data showed differences in service use between model 1 (larger, 24/7 services) and the other models. In the last month of life (0–14, 15–28 days), larger proportions of dyads in model 1 reported no visits into the home at all; use of out-of-home care was, however, greater than in the other three models. A total of 12% of dyads in model 1 reported no in-home care, no out-of-home care and no telephone calls between 0 and 14 days (7% between 15–28 days). Furthermore, relatively high proportions in model 1 reported no informal care (32% between 0 and 14 days, 18% between 15 and 28 days). Between 29 and 92 days, there continued to be smaller proportions of dyads in model 1 reporting in-home service use and greater proportions reporting out-of-home care than in the other three models, but all dyads reported use of at least one in-home, out-of-home or telephone contact, and all respondents reported informal care. Greater proportions of dyads in model 1 reported overnight stays (hospital, hospice or care home) during days 0–14, but not in the other time periods (see Appendix 20). As comparisons across models were standardised for time from death, a plausible reason for these differences between model 1 and the other models could not be identified.
Provision of nursing and care by hospice at home services
The overall numbers of home visits per day of community nurses/DNs and HCAs, nurses and HCAs from a HAH service and personal carers from social services were highest in model 2 (smaller, 24/7 services) and lowest in model 1 (larger, 24/7 services). There was no difference between the models with respect to the proportion of all nursing and personal caring visits that were made by community nurses/DNs in any of the three time periods (between 30% and 35% of all visits), but differences did exist between models in the proportions of HAH nurses or HCAs and social service personal carer visits. Pairwise comparisons of models showed that greater proportions of visits were from HAH staff in models 2 (smaller, 24/7 services) and 4 (smaller, not 24/7 services) than from HAH staff in models 1 (larger, 24/7 services) and 3 (larger, not 24/7 services); the proportion of visits from social service carers was greater in model 3 (larger, not 24/7 services) than in the other three models ( and ) (see also Appendix 22, –).
Summary statistics for total nursing and social care visits per day by different groups of local providers and time period
Comparisons of visits from HAH services and social services by model and time period
Costs
Costs are presented in GBP, at 2019 values. Full details of daily costs for all items of service use by time period and model are shown in Appendix 23, and for individual HAH services in Appendix 24, –. Summaries of costs by service model and time period for the main cost items (all nursing and personal care; all formal care, including nursing, personal and inpatient; informal care; grand total of formal and informal care) are shown in – and in Appendix 25.
Costs per service model: all care (formal and informal). Boxes show IQR with median bar inside. Whiskers are minimum to maximum, after excluding all outliers. Outliers (small circles) are 1.5–3 IQRs from the end of the box; extreme outliers (asterisks) (more...)
Costs per service model: informal care. Boxes show IQR with median bar inside. Whiskers are minimum to maximum, after excluding all outliers. Outliers (small circles) are 1.5–3 IQRs from the end of the box; extreme outliers (asterisks) are > 3 (more...)
Costs per day were higher closer to death: the median daily costs for 0–14, 15–28 and 29–92 days before death for all formal care were £104.57, £80.08 and £56.07, respectively. Among these overall costs, the median daily costs for in-home nursing and personal/social care were £40.43, £27.93 and £12.22 for 0–14, 15–28 and 29–92 days before death, respectively. Informal care costs, valued at replacement costs (support worker), exceed formal care costs, with daily medians of £580.00, £449.50 and £348.00 for 0–14, 15–28 and 29–92 days before death, respectively.
Statistically significant differences existed between models in all cost summary items and time periods (). Pairwise comparisons of models revealed that costs per day were consistently lower in model 1 (larger, 24/7 services) than in the other three models (for both formal and informal care). Daily costs in model 2 (smaller, 24/7 services) were higher than those in model 4 (smaller, not 24/7 services) for in-home nursing and personal/social care (all periods), but only for all formal care for days 29–92. Overall formal care costs were lower in model 4 (smaller, not 24/7 services) than in model 3 (larger, not 24/7 services). There were few differences in costs between model 2 (smaller 24/7 services) and model 3 (larger, not 24/7 services) (see ).
Comparison of costs per day (2019 GBP) between models for the main items of service use by model and time period
Summary and conclusions
Characteristics of participants
Differences existed between service models in the characteristics of patients recruited, with implications for the extent of data collection and interpretation of outcomes. Patients in model 1 (larger, 24/7 services) were in the study for longer (recruited further from death) than those in the other three models and had significantly better Australian modified Karnofsky Performance Status (functional status) scores at recruitment. There were similar differences in Phase of Illness scores at recruitment. The carers in models 1 and 4 were older than those in models 2 and 3.
There were also differences in patient characteristics between HAH services within service models that further impeded the interpretation of findings. In addition, some HAH services recruited small numbers, such that model outcomes reflected the outcomes from one larger service.
Primary outcome
The post-bereavement QODD was completed by 128 carers (response rate of 46.5%). Greater proportions of bereaved carers provided QODD responses in model 1; a higher education level was also associated with completion of the QODD. Across all models, the median QODD score (achieved by 52.3% of respondents) was 70.7, equivalent to a ‘good death’ [QODD scores range from 0 (worst) to 100, with > 70 being deemed ‘a good death’]. Higher QODD scores were elicited from female carers, from university-educated carers, when patients had known for a longer time that they were dying and when patients had died at home or in a hospice. When all items were adjusted for, models 2 and 4 (smaller providers, 24/7 and not 24/7 services, respectively) delivered significantly higher (around 12) QODD scores.
Preferred place of death
Of 222 patients who had expressed a PPOD (at recruitment) and for whom the actual place of death was known, 162 (73.0%) died in their preferred location [168 (75.7%) if care/nursing home deaths are counted as ‘home’]. There was no statistically significant difference between the four service models in achieving PPOD. Overall, 152 (68.5%) participants died at home [160 (71.5%) if deaths in care/nursing homes are included as home deaths]. Being in the study for a longer period of time was associated with a lower likelihood of achieving PPOD, possibly reflecting that preferences were gathered at recruitment and may change closer to death.
Satisfaction with care
Views on satisfaction with care were gathered from carers post bereavement, at the same time as completion of the QODD. Responses to VOICES 1 showed that most (76.4%) respondents thought that they had got as much help and support as they needed in the period before the patient died. The carers of female patients reported lower levels of support than the carers of male patients; carers in model 2 (smaller, 24/7 services) were eight times more likely to report getting all the help and support they needed than carers in the other three models. Most carers (75.8%) rated the help and support they received as excellent or outstanding (VOICES 2). Better ratings were associated with university-educated carers; the patient dying in hospital was associated with worse ratings. There was a trend for carers in model 2 to report better quality of support.
Service use
Rates of formal and informal care tended to increase closer to death, but there was considerable individual-level variation. During the last 2 weeks of life, carers reported medians per day of > 20 hours of informal caring, 1.76 in-home nursing and personal care visits and 2.36 total contacts (health, social and voluntary sectors, including nursing but excluding inpatient stays). Use of inpatient beds (hospital, hospice and care/nursing home) was low.
Role of hospice at home
Local NHS community services, HAH services and social services worked together to provide in-home nursing and care, and this was the most frequently reported item of formal service used. The highest daily average number of visits was reported in model 2 (smaller, 24/7 services) and the lowest in model 1 (larger, 24/7 services). There was no difference between the four models in the proportion of in-home nursing and care provided by local NHS community services (between 30% and 35% of all visits), but HAH and social services support appeared substitutable, with HAH input being greater in models 2 and 4 (smaller services).
Costs
Daily formal and informal care costs increased closer to death. Median daily total formal care costs were £104.57 in the last 2 weeks of life and £56.07 prior to the last month, of which in-home nursing and personal care daily costs were £40.43 and £12.22 in the last 2 weeks and prior to the last month, respectively. In line with reported service use, costs per day were consistently lower in model 1 (for formal and informal care) than in the other three models. Daily in-home nursing and care costs were higher in model 2 (smaller, 24/7 services) than in model 4 (smaller, not 24/7 services); the daily costs of all formal care were higher in model 3 (larger, not 24/7 services) than in model 4 (smaller, not 24/7 services). Informal care costs (which were valued by replacement cost methods) were higher than formal care costs in all models: the median daily costs (using replacement value) were £580.00 in the last 2 weeks of life and £348.00 beyond the last month.
