Does a palliative medicine service reduce hospital length of stay and costs in adults with a life-limiting illness?—a difference-in-differences evaluation of service expansion in Ireland

Introduction

Background

Improving the quality and value of care to people with incurable serious illness is a global policy priority, as part of efforts to reform health systems to meet the needs of ageing, multimorbid populations (1,2). There is widespread evidence of modifiable negative experiences in this population, including unmanaged pain and other symptoms, avoidable adverse events, and care inconsistent with preferences (3,4). These sub-optimal outcomes co-occur with high costs: people with incurable chronic illness account disproportionately for healthcare spending in high-income countries, and costs increase with proximity to death (5,6).

Palliative care is an approach that aims to improve quality of life and other outcomes for people with serious medical illness (7). In Ireland, palliative care services are organised into specialist and general palliative care services that operate in partnership as part of an integrated network of providers (8). In this paper, the term ‘palliative care’ is used as an umbrella term for services providing a palliative care approach and/or specialist palliative care services. The term ‘specialist palliative care’ is used to refer to services or people who provide specialist palliative care only.

In the acute hospital setting, the dominant model of specialist palliative care is palliative medicine service (PMS) consultation teams (9,10). The PMS becomes involved in patient care on the invitation of the attending physician and provide advice on the management of physical, emotional and/ or psychosocial distress, and initiate communication and goals-of-care discussions (11,12). Growing population health needs require expansion and adaptation of existing capacity (3,13). Clinical studies have found that palliative care for hospital in-patients improves outcomes (14), but the evidence on costs that is required to inform planning, value for money and optimal distribution of resources is distinctively small. While there has been a large number of observational cohort studies reporting reduced intensity of care (15,16), the number of studies providing high-quality evidence is far smaller (14,17,18). Primary research on palliative care faces practical and ethical challenges, resulting in a relatively small number of randomised trials of variable quality (14,17,19). Investigators have long relied on routinely collected data but evaluative studies face systematic selection bias and unobserved confounding (18,20).

Recent developments in economics, public health and other disciplines have advanced understanding on how credible causal estimates can be derived from routinely collected data under certain conditions, generating higher-quality evidence in the absence of randomised trials (21). The potential for such ‘quasi-experimental’ methods to improve economic evidence on end-of-life care is clear but relatively few studies have applied these methods to date (18).

Rationale and objectives

Palliative care services are well established in Ireland compared to similar countries and was among the first nations to publish a dedicated national policy in 2001 (13,22). Hospital care accounts for more than half of formal health care costs and proximity-to-death is a strong predictor of costs (23). Approximately half of the acute public hospitals in Ireland had some PMS involvement in inpatient care when the policy was published, and all other acute public hospitals have introduced PMS services such that by 2016 every hospital had some level of PMS involvement, albeit below recommended staffing levels in all hospitals (see Appendix 1 for more details) (8). Different hospitals implemented PMS teams at different times, providing the opportunity for a quasi-experimental evaluating using difference-in-differences analysis. We aimed to exploit this staggered PMS implementation to evaluate the intervention and derive credible causal estimates of its effects. Our primary aim was to evaluate how PMS implementation affected hospital inpatient length of stay (LOS). In secondary analysis we verified that observed LOS effects did not co-occur with changes in how many hospitalisations ended in death, and we estimated any cost differences associated with identified LOS effects. We present this article in accordance with the STROBE reporting checklist (available at https://apm.amegroups.com/article/view/10.21037/apm-23-479/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the ethics committee of the Trinity College Dublin Centre for Health Policy and Management (application number 05/2020/06E). Per the policies of the Healthcare Pricing Office (HPO) and the data-sharing agreement between HPO and the PI, individual consent for this retrospective analysis was waived.

Study design

This was a secondary analysis on routinely collected data for acute admissions to public hospitals in Ireland. A recent systematic review identified four quasi-experimental frameworks (18). We selected a difference-in-differences framework as appropriate in the context of our data. Difference-in-differences requires sequential, consistently recorded measures on outcome(s) of interest before and after the implementation of a clearly-defined intervention for group(s) that receive the intervention and group(s) that do not (24,25). Our data meet the criteria: acute inpatient LOS is recorded routinely in Irish public hospital data, and this collection occurred throughout the period of PMS expansion. Of the other frameworks, regression discontinuity was wholly unsuited to our data; interrupted time series analysis was inappropriate due to too few timepoints; and, consistent with many prior studies, we could not identify a suitable instrumental variable.

An overview of the difference-in-differences framework is provided in Figure 1.

Figure 1 Overview of difference-in-differences research design.

For further details of our methodological choices in the context of published guidance, see Appendix 2.

Hypothesised impact model

We hypothesised that PMS implementation would reduce LOS by expediting patient discharge and improving connection to community services including home care. We hypothesised that this reduction would not coincide with an increasing proportion of admissions ending in death, which might otherwise potentially explain reduced LOS.

Setting

Ireland has a population of approximately 5 million people and is relatively early in the demographic ageing process among high-income countries (26). As well as relatively strong palliative care services, notable health care system characteristics include weak primary care capacity and lack of universal primary care access; high numbers and turnover of acute inpatient hospital admissions; high medications spending; the largest duplicative voluntary insurance market in the world that can facilitate access based on ability to pay rather than medical need; and a fast rate of population ageing (13,27,28). Mean health care costs among people aged over 55 have been estimated as €8,053 (in 2022 euros), with marked increases towards end of life (12). There are 34 acute public hospitals with an emergency department in Ireland, which between them account for over 95% of acute inpatient admissions and over 90% of hospital inpatient deaths (29). The bulk of remaining admissions are to private hospitals covered by voluntary insurance and/or out-of-pocket payments (30), and these admissions are not included in this study.

Data sources

Our primary data source was the Hospital Inpatient Enquiry (HIPE) database, which records every admission to a public hospital in Ireland (31). Included categories of data include dates of admission and discharge, admission type (booked or emergency), sex, county of residence, diagnoses and procedures (using ICD-10 codes), LOS (days), and discharge destination (died in hospital, discharged home, hospice, nursing home, other hospital, other place). Due to the lack of a unique patient identifier in the Irish health care system, there is no way to quantify how many individuals are reflected in the data or to track their movements before or after admission (29,31).

We had multiple additional data sources. First, to define our exposure variable, we independently compiled a database of PMS implementation through communication with each of the 34 hospitals, identifying the year when a consultant-led PMS started (see Appendix 1). Second, to define our treatment variable, we used a database of PMS activity at one hospital that is linked locally to HIPE data (see Appendix 3). Third, to estimate costs associated with observed LOS, we used a unit cost of €938 per day in an Irish hospital based on prior work by others (32).

Sampling

Our sampling strategy is detailed in full in Appendix 1. Briefly, we first identified those hospitals that were eligible for the study based on hospital type (acute hospitals with an ED); available data on their PMS implementation; and not having had a PMS in 2009, the earliest point at which HIPE data were available. Second, we reviewed the characteristics of the HIPE data in those hospitals to assess suitability with respect to (I) using the ICD-10 code Z51.5, which we required to define our treatment variable; and (II) comparability of general outcome trends—those hospitals whose levels and/or trends in LOS and/or in-patient mortality rate differed substantively from the majority of sites were excluded due to comparability concerns. Third, having finalised our selection of sites and established the relevant timeframe for the analysis, we extracted data on HIPE observations (i.e., unique hospital in-patient admissions). For each site we first identified in the years following PMS implementation, those admissions with an ICD-10 code Z51.5 indicating palliative care provision. Within each site, we used propensity scores to identify the most similar observations in the years prior to PMS implementation (i.e., those most likely to have received palliative care had a service been in place). For each year at each site in the pre-PMS years we extracted the same number of observations as the number to receive palliative care in the year following implementation, thus creating a balanced panel with an equal number of observations per year at the hospital level.

Variables

Dependent variables

Our dependent variable in primary analysis was LOS, which was derived as continuous variable in days using admission and discharge dates. We conducted a secondary analysis on results using discharge disposition, which was derived as a binary variable (alive/dead) from the discharge destination variable. We estimated cost differences by combining estimated LOS differences in primary analysis with the unit cost of a hospital day in Ireland.

Independent variables

Our primary independent variable of interest was ‘did the person receive timely palliative care following hospital admission?’. We identified palliative care provision using the ICD-10 code Z51.5. In the context of prior research showing that timing of palliative care provision is essential to evaluating service effects on inpatient utilisation (33,34), we refined those observations with a palliative care interaction to those with a timely palliative care provision based on when the ICD-10 code was entered into the electronic health record. The process for defining the treatment variable is presented in Appendix 3; those classified as receiving palliative care on average received their first consultation within five days of admission. Additional predictors in calculating the propensity scores were age, sex, diagnosis of six specific conditions, and comorbidity total using the Elixhauser index (see Table 1) (37).

Statistical methods

Our difference-in-differences model is detailed in Appendix 2. Data supporting the common trends assumption are presented in Appendix 4. Primary analyses were conducted using Stata, version 15 (38). We report estimated average treatment effect on the treated (ATT), calculated using the margins command.

Propensity score weights were calculated in R using the cbps programme (39). Diagnostic assessment of matching on observed covariates was checked using absolute standardised difference and other diagnostic criteria (40). Each time we changed the sample (e.g., excluded a site, or changed the definition of treatment to include later consults), we recalculated the propensity scores.

Sensitivity analyses (SA)

We conducted the following SA:

• SA I. Without individual-level predictors on the right-hand side of the regression.
  • Rationale: selection of predictors is a debated point in difference-in-differences due to bias concerns (25). We retained predictors in primary analysis since each observation (admission) is unique and cohorts of observations therefore may vary across years on age or sex or diagnostic profile even after matching and we wanted to control for this variation while also checking it was not driving results.
• SA II. Without propensity score weights applied.
  • Rationale: propensity score weights may bias estimates particularly in contexts where the score is not balanced (25).
• SA III. Log-transformed LOS as the outcome of interest.
  • Rationale: standard difference-in-differences relies on a normality assumption, but health care utilisation data are skewed and heteroskedastic. We used untransformed LOS in primary analysis to support interpretation and in the context of the retransformation problem (41). SA III checks that this choice is not substantively affecting results.
• SA IV. By site.
  • Rationale: as detailed in Appendix 2, there is ongoing debates on appropriate ways to handle ‘always treated’ observations in multi-group, multi-year studies (25); as detailed in Table 1 and Appendix 4, one site (site 3) is distinctive in having small cell sizes. By reanalysing without site 3 and without always-treated observations we verify the robustness of primary results.
• SA V. By timing.
  • Rationale: as a sanity check on our treatment variable, we reran the primary analysis where the treatment group contained all PMS interactions (and not only those that occurred in a timely way) (42). The absolute magnitude of any ATT estimate ought to be smaller by construction, per prior research (33,34).

Results

Final sample and descriptive data

Our primary analytic sample included 4,314 unique admissions at four hospitals for the years 2009–2015; this constituted approximately 5% of all admissions at those sites during 2010–2015 (see Appendix 1). Descriptive statistics for the samples at each site are provided in Table 1. More than half of the observations were accounted for by one hospital (site 1). Differences in patient population were observable across sites; site 1 was a younger population with a high prevalence of cancer; sites 2 and 4 were older with lower prevalence of cancer and higher prevalence of circulatory diseases.

Outcome data

Prevalence of timely PMS provision, mean LOS and in-hospital mortality at each site are presented in Table 2. Prevalence of timely PMS provision is highest in site 1 (17%) and lowest in site 3 (6%), which primarily reflects implementation timing (PMS occurred earlier in sites 1 and 4 than in 2 and 3). Average LOS in our primary analytic sample was 10.6 days and a quarter of admissions ended in death.

Outcome data were examined for current trends assumptions prior to analysis, see Appendix 4.

Primary analysis

Our estimated ATT for timely palliative care at admission in four hospitals in Ireland between 2009 and 2015 is presented in Table 3. We found that timely PMS consultation reduced LOS by an estimated 1.94 days (95% confidence interval: −3.28 to −0.60).

Also presented in Table 3 are the results from SA detailed above. We found that our results were substantively robust to use of propensity score weights (SA I), use of observation-specific predictors (SA II), log-transforming the outcome variable (SA III) and excluded ‘always treated’ observations (SA IV). The estimated treatment effect did not persist once we reframed the treatment group eligibility as any PMS interaction irrespective of timing (SA V).

Secondary analysis

Results of our secondary analyses, first confirming that observed LOS differences are not co-occurring with discharge disposition differences, and second estimating the cost savings associated with reduced LOS in primary analysis, are presented in Table 4.

Discussion

Key results

We exploited the staggered implementation of PMS in acute public hospitals in Ireland to derive a credible causal estimate of the effect of timely palliative care following admission on LOS. We found that timely palliative care reduced LOS by nearly two days, with an estimated associated saving per admission of €1,820. These analyses were robust to multiple SA on regression specification, weighting strategy and site selection. It is the first natural experiment evaluation of hospital palliative care that we are aware of in a European context.

Limitations

The most significant limitations relate to data access and quality. Our ‘timely palliative care’ exposure variable is defined probabilistically, in the absence of deterministic linkage between HIPE and PMS databases (Appendix 3). Consequently we are unable to identify the ‘break point’ at which palliative care becomes cost-effective; this is a long-standing recognised issue in the field (43), and something that difference-in-differences can in principle address, with better data than were available to us. ICD codes for palliative care receipt have better specificity than sensitivity (44), i.e., we may have excluded incorrectly some post-PMS implementation observations where a palliative care interaction occurred but was not recorded. This ‘dose’ of specialist palliative care is sub-optimal since all services were understaffed against workforce planning recommendations, and the data don’t capture anywhere the value of generalist palliative care input (good generalist provision by primary teams may obviate need for specialist care; poor generalist provision may undermine the efforts of specialist teams).

While 34 acute public hospitals in Ireland had a PMS by 2016, only four of these were eligible for our study due to data quality and/or methodological limitations. The generalisability of results at these four specific hospitals to other parts of Ireland and internationally is unclear; for example, the best-quality evidence from Cochrane reviews suggests that hospital palliative care reduces risk of hospital death (14), but we did not find this relationship in our data. Our use of unit costs (instead of individual-level costs) increases the risk of error in treatment effect estimation, with possible bias in both directions. While prior evidence suggests that palliative care may reduce the intensity and length of a hospital stay (45,46), our cost data do not equip us to estimate savings related to reduced intensity, which may depress cost-effect estimates. We don’t have a reliable cost of palliative care provision for patients, and excluding this non-zero cost will inflate cost-estimates.

Our focus on the inpatient admission and formal costs maintains three limitations familiar in these studies: we don’t evaluate cost-effectiveness, i.e., quantifying how specialist palliative care may improve outcomes even when costs are unchanged; we don’t capture the unpaid family carer costs, which presumably increase when discharge is expedited (47); and we start from an implicit assumption that the inpatient admission was inevitable, whereas the largest cost-savings would accrue from better anticipatory care prior to admission that prevent hospital attendance.

Interpretation

Our findings have a number of important implications both for policy and for future research. For a literature characterised by large numbers of observational studies and few studies that derive credible causal treatment effect estimates, our results strengthen the evidence that timely hospital palliative care accelerates discharge for patients and saves costs to providers. The findings are broadly in line with other studies both using natural experiment and non-causal frameworks (15,16,18). Hospital costs account for more than half of formal costs among people with serious illness in Ireland, and hospital capacity is insufficient to meet current and growing needs. As such interventions that mitigate inpatient utilisation are a priority. Contingent on a simple assumption about the concurrent outcomes based on best available evidence (14), this suggests that timely hospital palliative care is cost-effective from the system perspective.

It is notable that only a minority of patients with a serious life-limiting illness admitted to hospital received palliative care once a PMS was in place, which implies that expanding PMS capacity would yield further gains. This capacity growth would first have to reach recommended staffing levels for need today, and then continue to grow in the context of rising needs (8,13). Also important, only a minority of those patients had their first interaction within the first few days of admission. When we expand our treatment group to all palliative care patients irrespective of treatment timing (SA V in Table 3), there is no statistically significant LOS reduction. Thus, while expanding PMS capacity may be important, ensuring that existing capacity is engaged early in an episode of care is perhaps as influential in optimising care for this patient population. In the Irish context, at least one strategy for the efficacious identification of hospital inpatients following emergency admission, has already been demonstrated (46).

Future research

This work offers a number of directions for follow-on work. Specifically in the case of natural experiments in evaluation of palliative care, we faced challenges in accessing quality data and in matching those data to a fast-evolving methodological landscape (25). This experience emphasises the value of careful identification of circumstances not only where new palliative care services have been implemented but also where high-quality individual data on the relevant populations. Such studies are critical to the long-term development of an evidence base where trials are relatively rare and will never generate sufficient evidence to address all questions.

More broadly, our finding that palliative care reduces hospital costs but only where the interaction occurs early, emphasises the need for more research on how to move the mean time of first interaction earlier in the admission. Such enquiries will surely encompass implementation science and evaluation of early identification strategies. PMS teams become involved only in the patient care at the invitation of attending teams, and so there is necessarily a mixed-methods element to these enquiries that can identify when and how attending teams would understand and welcome PMS input. Studies outside the hospital silo, both with respect to reducing inpatient admissions and capturing household perspectives, remain essential.

Conclusions

We evaluated the effect of hospital palliative care on utilisation in a natural experiment framework. We found that the service reduced LOS and costs provided the interaction occurs in a timely way following admission. Prompt interaction between suitable patients and palliative care can improve the quality and efficiency of care to this population. Future studies can extend and strengthen our approach with better data, as well as using different methods to understand how to trigger palliative care early in a hospital admission and realise available gains.

Acknowledgments

The PELCI project enjoyed the support of collaborators across the academic, public and NGO sectors. We are grateful for all support and participation.

Funding: This work was funded by the Health Research Board (Ireland) (Grant No. 2019/SDAP/012).

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Annals of Palliative Medicine for the series “Value of Palliative Care”. The article has undergone external peer review.

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://apm.amegroups.com/article/view/10.21037/apm-23-479/rc

Data Sharing Statement: Available at https://apm.amegroups.com/article/view/10.21037/apm-23-479/dss

Peer Review File: Available at https://apm.amegroups.com/article/view/10.21037/apm-23-479/prf

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://apm.amegroups.com/article/view/10.21037/apm-23-479/coif). The series “Value of Palliative Care” was commissioned by the editorial office without any funding or sponsorship. P.M. served as the unpaid Guest Editor of the series. This work is part of a project titled ‘Palliative and end-of-life care data in Ireland: establishing the state of the nation, mapping future direction’, which was funded through the Irish Health Research Board 2019/SDAP/012 [PI: May]. The PELCI project enjoyed the support of collaborators across the academic, public and NGO sectors. The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the ethics committee of the Trinity College Dublin Centre for Health Policy and Management (application number 05/2020/06E). Per the policies of the Healthcare Pricing Office (HPO) and the data-sharing agreement between HPO and the PI, individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Knaul FM, Bhadelia A, Rodriguez NM, et al. The Lancet Commission on Palliative Care and Pain Relief—findings, recommendations, and future directions. The Lancet Global Health 2018;6:S5-6. [Crossref]
2. European Commission. Ageing report economic and budgetary projections for the 28 EU Member States (2013-2060). Luxembourg; 2015.
3. Sleeman KE, de Brito M, Etkind S, et al. The escalating global burden of serious health-related suffering: projections to 2060 by world regions, age groups, and health conditions. Lancet Glob Health 2019;7:e883-92. [Crossref] [PubMed]
4. Sallnow L, Smith R, Ahmedzai SH, et al. Report of the Lancet Commission on the Value of Death: bringing death back into life. Lancet 2022;399:837-84. [Crossref] [PubMed]
5. French EB, McCauley J, Aragon M, et al. End-Of-Life Medical Spending In Last Twelve Months Of Life Is Lower Than Previously Reported. Health Aff (Millwood) 2017;36:1211-7. [Crossref] [PubMed]
6. Breyer F, Lorenz N. The "red herring" after 20 years: ageing and health care expenditures. Eur J Health Econ 2021;22:661-7. [Crossref] [PubMed]
7. World Health Organization. WHO Definition of Palliative Care 2017. Available online: http://www.who.int/cancer/palliative/definition/en/.
8. Ryan K, Creedon B, Myers K, Marsden M, Reaper-Reynolds S. National Clinical Programme for Palliative Care Model of Care. 2019. Available online: https://www.hse.ie/eng/about/who/cspd/ncps/palliative-care/moc/
9. Morrison RS. Models of palliative care delivery in the United States. Curr Opin Support Palliat Care 2013;7:201-6. [Crossref] [PubMed]
10. Wee B. Models of delivering palliative and end-of-life care in the UK. Curr Opin Support Palliat Care 2013;7:195-200. [Crossref] [PubMed]
11. Brereton L, Clark J, Ingleton C, et al. What do we know about different models of providing palliative care? Findings from a systematic review of reviews. Palliat Med 2017;31:781-97. [Crossref] [PubMed]
12. Health Service Executive. Palliative Care Services: Three-year development framework. Dublin: HSE Primary Care Division; 2017.
13. May P, Johnston BM, Normand C, et al. Population-based palliative care planning in Ireland: how many people will live and die with serious illness to 2046? HRB Open Res 2019;2:35. [Crossref] [PubMed]
14. Bajwah S, Oluyase AO, Yi D, et al. The effectiveness and cost-effectiveness of hospital-based specialist palliative care for adults with advanced illness and their caregivers. Cochrane Database Syst Rev 2020;9:CD012780. [PubMed]
15. Smith S, Brick A, O'Hara S, et al. Evidence on the cost and cost-effectiveness of palliative care: a literature review. Palliat Med 2014;28:130-50. [Crossref] [PubMed]
16. Luta X, Ottino B, Hall P, et al. Evidence on the economic value of end-of-life and palliative care interventions: a narrative review of reviews. BMC Palliat Care 2021;20:89. [Crossref] [PubMed]
17. Gomes B, Calanzani N, Curiale V, et al. Effectiveness and cost-effectiveness of home palliative care services for adults with advanced illness and their caregivers. Cochrane Database Syst Rev 2013;2013:CD007760. [PubMed]
18. Jiang J, Kim N, Garrido MM, et al. Effectiveness and cost-effectiveness of palliative care in natural experiments: a systematic review. BMJ Support Palliat Care 2023;spcare-2022-003993.
19. Higginson IJ, Evans CJ, Grande G, et al. Evaluating complex interventions in end of life care: the MORECare statement on good practice generated by a synthesis of transparent expert consultations and systematic reviews. BMC Med 2013;11:111. [Crossref] [PubMed]
20. Starks H, Diehr P, Curtis JR. The challenge of selection bias and confounding in palliative care research. J Palliat Med 2009;12:181-7. [Crossref] [PubMed]
21. Khullar D, Jena AB. “Natural Experiments” in Health Care Research. JAMA Health Forum 2021;2:e210290. [Crossref] [PubMed]
22. Arias-Casais N, Garralda E, Rhee J, et al. EAPC Atlas of Palliative Care in Europe 2019. Vilvoorde: EAPC Press, 2019.
23. May P, Moriarty F, Hurley E, et al. Formal health care costs among older people in Ireland: methods and estimates using The Irish Longitudinal Study on Ageing (TILDA). HRB Open Res 2023;6:16. [Crossref] [PubMed]
24. Jagielka P. Impact Evaluations [Internet]. Washington, DC: World Bank. 2019. Available online: https://blogs.worldbank.org/impactevaluations/what-are-we-estimating-when-we-estimate-difference-differences.
25. Roth J, Sant'Anna PHC, Bilinski A, Poe J. What's Trending in Difference-in-Differences? A Synthesis of the Recent Econometrics Literature. Journal of Econometrics 2023;235:2218-44. [Crossref]
26. Kane PM, Daveson BA, Ryan K, et al. The need for palliative care in Ireland: a population-based estimate of palliative care using routine mortality data, inclusive of nonmalignant conditions. J Pain Symptom Manage 2015;49:726-733.e1. [Crossref] [PubMed]
27. Organisation for Economic Co-operation and Development, Systems EOoH, Policies. Ireland: Country Profile. Paris2017.
28. Nolan A, Ma Y, Moore P. Changes in Public Healthcare Entitlement and Healthcare Utilisation among the Older Population in Ireland. Dublin: The Irish Longitudinal Study on Ageing; 2016.
29. Matthews S, Pierce M, O'Brien Green S, Hurley E, Johnston B, Normand C, et al. Dying and Death in Ireland: What Do We Routinely Measure, How Can We Improve? Dublin: Irish Hospice Foundation; 2021.
30. Connolly S, Wren M-A. Universal Health Care in Ireland—What Are the Prospects for Reform? Health Syst Reform 2019;5:94-9. [Crossref] [PubMed]
31. Healthcare Pricing Office. HIPE Data Dublin 2021. Available online: http://www.hpo.ie/.
32. Carter L, Yadav A, O'Neill S, et al. Extended length of stay and related costs associated with dementia in acute care hospitals in Ireland. Aging Ment Health 2023;27:911-20. [Crossref] [PubMed]
33. May P, Garrido MM, Cassel JB, et al. Using Length of Stay to Control for Unobserved Heterogeneity When Estimating Treatment Effect on Hospital Costs with Observational Data: Issues of Reliability, Robustness, and Usefulness. Health Serv Res 2016;51:2020-43. [Crossref] [PubMed]
34. May P, Normand C. Analyzing the Impact of Palliative Care Interventions on Cost of Hospitalization: Practical Guidance for Choice of Dependent Variable. J Pain Symptom Manage 2016;52:100-6. [Crossref] [PubMed]
35. Garrido MM. Propensity scores: a practical method for assessing treatment effects in pain and symptom management research. J Pain Symptom Manage 2014;48:711-8. [Crossref] [PubMed]
36. StataCorp. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC; 2017.
37. Imai K, Ratkovic M. Covariate balancing propensity score. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 2014;76:243-63. [Crossref]
38. Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med 2015;34:3661-79. [Crossref] [PubMed]
39. Manning WG. The logged dependent variable, heteroscedasticity, and the retransformation problem. J Health Econ 1998;17:283-95. [Crossref] [PubMed]
40. May P, Normand C, Noreika D, et al. Using predicted length of stay to define treatment and model costs in hospitalized adults with serious illness: an evaluation of palliative care. Health Econ Rev 2021;11:38. [Crossref] [PubMed]
41. Etkind SN, Bone AE, Gomes B, et al. How many people will need palliative care in 2040? Past trends, future projections and implications for services. BMC Med 2017;15:102. [Crossref] [PubMed]
42. Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measures for use with administrative data. Med Care 1998;36:8-27. [Crossref] [PubMed]
43. May P, Garrido MM, Cassel JB, et al. Prospective Cohort Study of Hospital Palliative Care Teams for Inpatients With Advanced Cancer: Earlier Consultation Is Associated With Larger Cost-Saving Effect. J Clin Oncol 2015;33:2745-52. [Crossref] [PubMed]
44. Hua M, Li G, Clancy C, et al. Validation of the V66.7 Code for Palliative Care Consultation in a Single Academic Medical Center. J Palliat Med 2017;20:372-7. [Crossref] [PubMed]
45. May P, Garrido MM, Cassel JB, et al. Cost analysis of a prospective multi-site cohort study of palliative care consultation teams for adults with advanced cancer: Where do cost-savings come from? Palliat Med 2017;31:378-86. [Crossref] [PubMed]
46. Tiernan E, Ryan J, Casey M, et al. A quasi-experimental evaluation of an intervention to increase palliative medicine referral in the emergency department. J Health Serv Res Policy 2019;24:155-63. [Crossref] [PubMed]
47. Gardiner C, Brereton L, Frey R, et al. Exploring the financial impact of caring for family members receiving palliative and end-of-life care: a systematic review of the literature. Palliat Med 2014;28:375-90. [Crossref] [PubMed]