Introduction

Good-quality care is needed for the ageing populations in the European Union (EU).[1, 2] While many countries aim to enable people to continue living in their own home, need of care in long-term care facilities (LTCFs) is growing.[3] This trend is costly for health care systems, and demands the assessment of efficiency of these facilities. Therefore, this study provides insight into variability in efficiency of LTCFs within and between six EU countries (Belgium–Flanders only, England, Finland, Italy, the Netherlands and Poland).

The ageing of our societies is substantially changing healthcare needs [4] and end-of-life (EoL) issues, and gives rise to concerns about quality of life (QoL) and quality of dying (QoD). Since a significant proportion of people live and eventually die in LTCFs, [5, 6], enabling people to live and die well grows in importance.[3] At the same time, long-term care delivery will continue driving up public spending,[7] as LTCFs in the EU are mostly funded out of public money.[7] Illustrating the cost issue: the proportion of long-term care costs in total healthcare expenditure currently are around 40% in the Netherlands and in Finland.[8] A recent FP7 European project (ANCIEN) has shown that a significant share of public expenditures is devoted to institutional care.[9] This trend is a major concern for most governments,[7] and the financing of EoL care in LTCFs will be up for discussion as these issues increasingly affect more people. The World Health Report 2000 called attention to the importance of efficiency in health systems and in achieving the goals of health improvement, responsiveness and fairness in financing.[10] The (balance between) quality of care and efficiency in LTCFs is a growing concern in many industrialized countries.

Thus, it is crucial to internationally compare the efficiency of LTCFs in different EU countries, and to identify factors affecting efficiency and quality of care. Yet, only few single country analyses have been performed in the EU,[11, 12] none of which specifically concerning the EoL. Björkgren et al. found that larger units were seemingly more efficient than smaller ones.[11] Garavaglia and colleagues indicated that private nursing homes appear to outperform public ones.[12] Other studies from Canada and the U.S. primarily focused on ownership as efficiency predictor. Lee et al., for example, found that for-profit facilities were more efficient than not-for-profit ones.[13]

The aim of our study was to investigate and explain variation in the technical efficiency of EoL care within and between LTCFs throughout six EU countries participating in the EU funded 7^th Framework Program ‘PACE’ (Palliative Care for Older People in Care and Nursing Homes in Europe).[3] Technical efficiency in this study reflects the LTCFs ability to obtain maximal output in terms of of QoL and QoD from a given input of personnel and capacity.

Methods

Production function

A data envelopment analysis (DEA) as method of technical efficiency analysis was chosen in this study since in this type of analysis, the efficiency of so-called decision making units (DMU’s) can be calculated. Efficiency scores of 1 indicate maximal efficiency. DMU’s with a score of 1 are used as benchmark for other participating LTCFs. Scores above 1 are considered less efficient. For example; an efficiency score of 1.3 indicates that it is possible to ‘produce’ 30% more output with the same inputs. Moreover, DEA can deal with multiple inputs and outputs, and needs no assumptions about the distribution between them.

In our DEA, LTCFs were handled as DMU’s where output is produced using a given number of resource inputs. The two-stage approach as suggested by Coelli et al. was adopted,[14] in which the first stage analysis involves solving a DEA problem (including only traditional inputs and outputs), and in the second stage efficiency scores from the first stage are regressed upon environmental variables. A quality-driven production function was conducted.[15] See Fig 1 for an illustration of our production function (stage 1), producing efficiency scores, and regression analysis (stage 2) on these scores. The selection of input, output, case mix and explanatory variables (see below) was based on literature as well as the expertise of PACE consortium members.[11, 12]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Production function.

https://doi.org/10.1371/journal.pone.0204120.g001

Input variables.

Data regarding LTCFs fulltime-equivalents (FTEs) of nurses–both registered and licensed practical, care assistants and allied health professionals (paramedics) were used. These FTEs were divided by the number of beds within LTCFs (indirectly adjusting for case mix), which number subsequently was multiplied by the occupancy rate. In this way capacity was also taken into account. GP visits were used as a proxy for the GPs involvement (since FTEs for GPs were not available in all countries as in most LTCFs they are not part of the permanent staff). Our approach was similar to that of Björkgren et al. and Garavaglia et al. by taking into account health, nursing and accommodation (capacity) as resource input.[11, 12]

Output variables.

Since in LTCFs, (maintaining) QoL and QoD are the most important outcomes, these were chosen as output variables. With regard to QoL, the EuroQol-5D (EQ5D) was used since this is the most often used outcome measure in economic evaluations in health care. The EQ5D is a standardized instrument measuring five health domains (mobility, self-care, usual activities, pain/discomfort and anxiety/depression).[16] In order to achieve utility scores–ranging from 0 (death) to 1 (perfect health)–for the EQ5D, value sets from England were used,[17] since value sets were not yet available for all participating countries.[18] However, the EQ5D arguably is not the most relevant tool for measuring QoL in EoL care, since it takes into account health related domains which are considered less relevant than other dimensions of QoL such as peace, emotions, spiritual and psychosocial well-being.[19] Therefore, two QoD outcome measures were additionally used as output variables. The Comfort Assessment in Dying with Dementia (EOLD-CAD), with four subscales: physical distress, dying symptoms, emotional distress, and well-being,[20] and the Quality of Dying in Long-Term Care (QOD-LTC), comprising of three subscales: personhood, closure, and preparatory tasks.[21] In both tools, higher mean scores reflect a higher quality of end-of-life. Both tools were rated by the nurse or care assistant most closely involved in care of the residents. Literature was consulted and potential output variables were discussed with experts of the PACE consortium. See Table 1 for all input and output variables, and their level of measurement.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Input and output variables.

https://doi.org/10.1371/journal.pone.0204120.t001

Stage 2. Truncated regression analysis

In order to explain differences in (bias corrected) efficiency scores from stage 1, truncated regression analysis was performed on a preselected set of explanatory and case mix adjusting variables. Belgium (Flanders) was used as reference country in the regression analysis.

Explanatory and case mix variables.

Country, status (ownership) and availability of palliative care and opioids were chosen as explanatory variables (see Table 2). Moreover, as technical efficiency differences of LTCFs were conditional on case mix, we made the comparison more homogeneous by taking into account the seven-item Bedford Alzheimer nursing-severity scale score (BANS-S, a rating scale (range 7–28) comprising cognitive and functional items) of deceased residents, percentage of residents needing assistance in eating, and the mean length of stay of residents in the LTCFs were used (see Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Explanatory variables.

https://doi.org/10.1371/journal.pone.0204120.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Case mix variables.

https://doi.org/10.1371/journal.pone.0204120.t003

Results

Of the 322 participating LTCFs, 176 (54,6%) had complete cases on all in- and output variables. After deleting LTCFs with missing values on explanatory and case mix variables, 173 LTCFs (53,7%) remained. PACE questionnaires regarding facility characteristics were completed for 94,7% LTCFs, response rates for questionnaires regarding the resident filled in by the staff ranged from 95,1% in Finland to 54,2% in England.

Inferential statistics

Of the 305 LTCFs, 176 (58%) could be taken into account in the stage 1 analysis. Because of heterogeneity in disease severity and QoL and QoD of residents per LTCF type (1, 2 and 3), in the base case analyses another 41 LTCFs were excluded because of not being type 2. Eventually, 135 LTCFs (34%) were taken into account in the base case analysis.

Stage 1. Output-oriented DEA

The stage 1 DEA showed that 48 out of 135 LTCFs (35%) were on the efficient frontier, meaning these LTCFs served as benchmarks for the other LTCFs. The results point to a significant difference from the efficient frontier (p < 0.01, mean efficiency score 1.56, 95%-CI: 1.45–1.66), indicating that–based on our production function–there is room for efficiency improvement. Next to this variation across all LTCFs, a significant difference in efficiency was found within countries (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Bias-corrected efficiency scores per country (the closer to ‘1’, the more efficient).

https://doi.org/10.1371/journal.pone.0204120.t004

STAGE 2. Truncated regression analysis–base case

After having obtained the bias-corrected efficiency scores in stage 1, truncated regression analysis was performed in order to explain differences in the scores. Due to missing values on explanatory variables, 133 LTCFs (out of the 135 in stage 1 remaining) were taken into account in the stage 2 analysis.

The regression analysis indicated considerable variability in LTCF efficiency throughout Europe (Table 5). The variation between countries however, did not differ significantly from benchmark country Belgium. Similarly, ownership, availability and/or use of palliative care or opioids did not influence efficiency.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Regression analysis (base case–all input, type 2 LTCFs).

https://doi.org/10.1371/journal.pone.0204120.t005

Scenario analysis.

When conducting the hands-on care at the bedside analysis (for type 2 LTCFs), Poland (p = 0.00) and Finland (p = 0.04) seemed to be more efficient than the benchmark Belgium. Again, ownership, availability and/or use of palliative care or opioids did not influence efficiency (Table 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Regression analysis (Hands-on care–nursing and care–at the bedside, type 2 LTCFs).

https://doi.org/10.1371/journal.pone.0204120.t006

Discussion

In this study, a quality-driven production function consisting of personnel (nurses, care assistants, allied health professionals/paramedics and GPs) and capacity (bed occupancy) as input, and QoL (EQ5D) and QoD (EOLD-CAD and QOD-LTC)–as judged by staff—as output was built to calculate technical efficiency scores for each LTCF. Country, type, ownership and the availability of palliative care and opioids were pre-selected to explain variability in efficiency. A base case analysis and a scenario analysis were conducted. A lot of variation in technical efficiency of partaking LTCFs was found, which persisted when looking within countries.

Both a base case analysis as well as a hands-on care at the bedside scenario were conducted. The latter as nurses and care assistant deliver the large majority of care in LTCFs,[24, 25] and they thus particularly are important in producing QoL and QoD. In order to improve heterogeneity, only type 2 LTCFs–with onsite care from nurses and care from offsite based GPs–were taken into account in our analyses. Efficiency scores suggest the English LTCFs tend to be more efficient, but this difference was not statistically significant. Looking at the included LTCFs from England in more detail, we observed they have a distinctive model characterized by relatively few nurses, many care assistants, and an emphasis on palliative care). Variation in efficiency did significantly differ between countries in the hands-on care at the bedside scenario, in which Poland and Finland seemed most efficient. In Poland, relatively good QoL and QoD was achieved despite low–possibly even understaffed–resource input. Moreover, it was noteworthy that in Finland the ‘hands-on care at the bedside’ is relatively high educated. I.e. the ratio nurses (registered and licensed practical) over care assistants is higher than in other participating countries. In this study, however, we only studied whether the LTCFs’ country is influencing efficiency, not why certain countries are more efficient. Future analyses of contextual frameworks would be important to explain further differences in efficiency.

Only resident-related data concerning the last weeks of life was used, making the group very homogeneous in terms of phase of life. However, variation in disease severity and closely related LTCF types was found, which can hardly be standardized. This was a disadvantage, since comparative research, especially DEA, is best undertaken with homogeneous data. Moreover, output (QoL & QoD) and disease severity (BANS-S) data were gathered among deceased residents, while at the same time input data (FTE) was based on the entire LTCF. Ideally, data of all LTCF residents would have been considered. Also as this gives insight in the comparability of LTCFs as a whole. This is challenging however, as it requires prospective data collection.

Moreover, healthcare professionals not having an appointment in the facility, who visit as required, were missed in our analyses although they and family or volunteers might also influence QoL and QoD of residents. Another limitation regards the fact that questionnaires regarding the deceased resident were filled in by a care provider from the LTCF staff. The knowledge level and understanding of questions–i.a. regarding the residents’ QoL and QoD–by the care provider might have influenced their answers; the use of outcomes judged by the staff might have led to an overestimation of QoL and QoD in countries where education in palliative care in LTCFs is low. Moreover, value sets from England were used to derive EQ5D utilities, as they are not yet available for all participating countries.[17] It would be valuable for future studies to be able using population-specific valuation sets better reflecting cultural differences.[26]

This pioneering large-scale study provides a first insight into the technical efficiency of end-of-life care delivery in LTCFs in six European countries. In terms of methodology our study is comparable to others.[11, 12] However, it is the first to make a comparison specifically in EoL care and across countries. Also, the fact that we did not follow the standard choice of using quantity (volumes) as output, but used quality is a strength.[27] Comprehensive comparative EU-wide research remains a challenge as it is influenced by many factors.

Supporting information

Acknowledgments

We thank all PACE collaborators (lead author: Lieve van den Block, LvdBlock@vub.ac.be) not in the author list: Zeger De Groote, Yuliana Gatsolaeva, Lara Pivodic, Elisabeth Honinx, Nele Van Den Noortgate, Ruth Piers, Marc Tanghe, Hein van Hout, Jenny T. van der Steen, Maud ten Koppel, Mariska Ooosterveld, Federica Mammarella, Martina Mercuri, Paola Rossi, Agnieszka Pac, Violetta Kijowska, Agata Stodolska, Harriet Finne-Soveri, Paula Andreasen, Outi Kuitunen-Kaija, Suvi Leppäaho, Katheine Froggatt, Danni Collingridge Moore, and the European Association for Palliative Care vzw, European Forum For Primary Care, Age Platform Europe, and Alzheimer Europe.