Open access Original research Provision of critical care for the elderly in Europe: a retrospective comparison of national healthcare frameworks in intensive care units Bernhard Wernly ‍ ‍,1,2 Michael Beil,3 Raphael Romano Bruno ‍ ‍,4 Stephan Binnebössel,4 Malte Kelm ‍ ‍,4 Sviri Sigal,3 Peter Vernon van Heerden,5 Ariane Boumendil,6 Antonio Artigas,7 Maurizio Cecconi,8 Brian Marsh,9 Rui Moreno,10 Sandra Oeyen,11 Bernardo Bollen Pinto,12 Wojciech Szczeklik ‍ ‍,13 Susannah Leaver,14 Sten Mikael Walther,15,16 Joerg C Schefold,17 Michael Joannidis,18 Jesper Fjølner,19 Tilemachos Zafeiridis,20 Dylan de Lange,21 Bertrand Guidet,22 Hans Flaatten,23 Christian Jung,4 On behalf of the VIP2 study group To cite: Wernly B, Beil M, ABSTRACT Bruno RR, et al. Provision of Objectives In Europe, there is a distinction between two Strengths and limitations of this study critical care for the elderly different healthcare organisation systems, the tax-­based in Europe: a retrospective ►► We performed a retrospective post hoc analysis of a healthcare system (THS) and the social health insurance comparison of national European cohort of critically ill patients who are 80 system (SHI). Our aim was to investigate whether the healthcare frameworks in years and older. intensive care units. BMJ Open characteristics, treatment and mortality of older, critically ►► This study evaluated critically ill older (≥80 years) 2021;11:e046909. doi:10.1136/ ill patients in the intensive care unit (ICU) differed between patients, 4941 in tax-­based healthcare system and bmjopen-2020-046909 THS and SHI. 2876 in the social health insurance system systems Setting ICUs in 16 European countries. ►► Prepublication history and from the VIP1 and VIP2 study. Participants In total, 7817 critically ill older (≥80 years) additional supplemental material ►► We compared baseline risk distribution using uni- patients were included in this study, 4941 in THS and 2876 for this paper are available variate methods according to the distribution. online. To view these files, in the SHI systems. ►► We compared management and mortality outcomes please visit the journal online Primary and secondary outcomes measures We chose using generalised estimation equations with robust (http://​dx.​doi.​org/​10.​1136/​ generalised estimation equations with robust standard standard errors to produce population average ad- bmjopen-​2020-​046909). errors to produce population average adjusted OR (aOR). justed aOR. We adjusted for patient-­specific variables, health economic Received 20 November 2020 ►► We performed multiple sensitivity analyses stratify- data, including gross domestic product (GDP) and human Accepted 13 May 2021 ing 30-­day mortality for patient-­specific characteris- development index (HDI), and treatment strategies. tics, treatment strategies and health economic data. Results In SHI systems, there were higher rates of frail patients (Clinical Frailty Scale>4; 46% vs 41%; p<0.001), longer length of ICU stays (90±162 vs 72±134 hours; p<0.001) and increased levels of organ support. The INTRODUCTION ICU mortality (aOR 1.50, 95% CI 1.09 to 2.06; p=0.01) Very old patients (often defined as being 80 was consistently higher in the SHI; however, the 30-­day years of age or older) admitted to an inten- mortality (aOR 0.89, 95% CI 0.66 to 1.21; p=0.47) was sive care unit (ICU) have a high mortality.1 similar between THS and SHI. In a sensitivity analysis Whether this group of patients will benefit stratifying for the health economic data, the 30-­day from intensive care treatment is the subject of © Author(s) (or their mortality was higher in SHI, in low GDP per capita (aOR clinical studies and ethical debate.2 3 Never- employer(s)) 2021. Re-­use 2.17, 95% CI 1.42 to 3.58) and low HDI (aOR 1.22, 95% CI theless, older patients consume a dispropor- permitted under CC BY-­NC. No 1.64 to 2.20) settings. tionate amount of intensive care resources, commercial re-­use. See rights Conclusions The 30-­day mortality was similar in both and permissions. Published by systems. Patients in SHI were older, sicker and frailer with questionable benefits on morbidity and BMJ. at baseline, which could be interpreted as a sign for a mortality, which raises both socioeconomic For numbered affiliations see more liberal admission policy in SHI. We believe that and ethical questions of sustainability.4–6 end of article. the observed trend towards ICU excess mortality in SHI There are numerous factors that influence results mainly from a more liberal admission policy and an outcome, age being only one of many.7–9 In Correspondence to Dr Christian Jung; increase in treatment limitations. addition to the severity of organ dysfunction, ​christian.​jung@​med.​uni-​ Trial registration numbers NCT03134807 and the patient’s functional and mental capacity duesseldorf.​de NCT03370692. prior to admission is of utmost importance.10 Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 1 Open access Multiple studies have confirmed the concept of frailty as February 2017. For VIP2, VIPs were included from May a valuable tool, not only to assess the patient’s prehospital 2018 to May 2019. Many countries could recruit patients condition, but also to predict outcome.11–14 without informed consent while some had to collect In addition to patient-­specific factors, specific health- informed consent as the ethical consent procedures in care system characteristics, such as availability of resources, Europe are diverse. For each patient baseline character- are hypothesised to influence the outcome of critically ill istics (including age, gender, main reason for admission older patients.15 In Europe, a rough distinction is made and frailty) and management strategies (including use between two healthcare systems; the tax-­ based health- of renal replacement therapy (RRT), mechanical venti- care system (THS) and the social health insurance system lation, non-­invasive ventilation (NIV) and rates of use of (SHI). There are other more nuanced classification vasoactive drugs) were documented. Also, any treatment systems with higher granularity used to divide the organ- limitations (treatment withheld, or treatment withdrawn) isation of healthcare systems.16–20 However, ultimately were documented. all subclassifications are variations of the initial subdivi- In this post hoc analysis of these two prospective trials, sion into a primarily tax-­ financed, municipally organ- all patients admitted acutely (ie, non-­ electively) with ised system with a ‘single payer’ (THS) or into a social complete data on age, gender, Clinical Frailty Score (CFS) insurance system financed by predefined and earmarked and Sequential Organ Failure Assessment (SOFA) Score premiums, mostly from payroll taxes, and organised in a and ICU mortality were included.9 Patients were defined more pluralistic way, often with the coexistence of both as being in a THS (Denmark, Great Britain, Sweden, private and public service providers.21 A dichotomisation Norway, Ireland, Italy, Spain, Portugal and Greece; of the European healthcare system into THS versus SHI n=4941) or SHI (Poland, Belgium, France, Germany, is therefore possibly an over-­simplification, but it is never- Netherlands, Austria and Switzerland; n=2876) health- theless based on distinct differences.17 Other pragmatic care system. We used the dichotomisation used by the analyses also use the same distinction, as well as the World World Bank and the OEEC.22 The primary endpoint of Bank and the Organisation for European Economic this study was ICU mortality, and the secondary endpoints Co-­operation (OEEC).22 Differences in the treatment of were 30-­day mortality and rates of treatment limitation. older people between these two systems are likely, as the The data on 30-­day mortality was available for only 7443 incentives for treating critically ill patients might differ patients (THS n=4712; SHI n=2731). Frailty was assessed between THS and SHI. In addition, differences in the by the CFS and the respective visual and simple descrip- way finances are allocated for elderly patients have been tions were used with permission.25–27 For the patients described.23 recruited to the VIP2 trial, Katz activities of daily living The VIP1 study evaluated 5021 critically ill patients (Katz ADL) with ADL Score≤4 defining disability and over the age of 80 and showed that frailty assessed by the short form of Informant Questionnaire on Cognitive Clinical Frailty Scale (CFS) was associated with short-­term Decline in the Elderly (IQCODE), with IQCODE ≥3.5 mortality. The VIP2 study included acutely admitted 3920 describing cognitive decline and Co-­morbidity and Poly- critically ill patients over the age of 79 and confirmed the pharmacy Score were assessed.25–27 The gross domestic association of CFS with short-­term mortality.14 24 product (GDP) per capita for 2019 in US$ was retrieved In this post hoc substudy of these two studies in very old from the International Money Fund,28 the human devel- intensive care patients (VIPs studies), we wanted to inves- opment index (HDI) from the United Nations Develop- tigate whether the characteristics, management strategies ment Program,29 and the total (compulsory, voluntary, (with a focus on organ support and treatment limitations) out-­of-­pocket) amount of health spending per capita in and mortality of older, critically ill patients in the ICU US$ in 2019 from the Organisation for Economic Co-­op- differed between THS and SHI.14 24 Since older patients eration and Development.30 The number of ICU beds in particular have a high probability of death, even after were obtained from a previous publication.31 survival of the initial intensive care stay, we have evaluated not only the ICU mortality but also the 30-­day mortality.3 Patient and public involvement We hypothesised that patients differ significantly between No patient involved. THS and SHI with regards to baseline characteristics due to differing admission policies. Therefore, we speculated Statistical analysis whether there were significant differences in the use of We assessed continuous variables for distribution using organ support and mortality rates between THS and SHI. Kolmogorov–Smirnov test. Normally distributed contin- uous data are given as mean±SD and compared using Student’s t-­ test. Non-­normally distributed continuous METHODS data are given as median±IQR and compared using Study subjects Mann’s U-­ test. Categorical data are given as numbers VIP1 and VIP2 were prospective, multicentre studies, regis- (percentage) and compared using the χ test. tered on ​ClinicalTrials.​gov.14 24 Both studies included very The primary exposure was being in a THS versus SHI. The elderly VIPs, defined as patients aged 80 years or older. data are likely to be clustered on an ICU level. Therefore, For VIP1, data were collected between October 2016 and a multilevel regression approach was chosen. As the health 2 Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 Open access Figure 3 Sensitivity analyses stratifying 30-­day mortality for Figure 1 Sensitivity analyses stratifying 30-­day mortality for health economic data using generalised estimation equations patient-­specific characteristics using generalised estimation producing population average ORs. The depicted aORs from equations producing population average ORs. The depicted model-1 include only the intensive care unit (ICU) as panel. aORs from model-1 include only the intensive care unit (ICU) The 30-­day-­mortality was higher in SHI, in settings with low as panel. The 30-­day mortality was similar between tax-­ ICU beds per capita (aOR 2.17 95%CI 1.57-3.02), low gross based healthcare system (THS) and health insurance system domestic product (GDP) per capita (aOR 2.17 95% CI 1.42- (SHI) in male (aOR 1.26, 95% CI 0.98 to 1.61), frail (aOR 1.15, 3.58) and low human development index (HDI) (aOR 1.22 95% CI 0.998 to 1.710), non-­frail (aOR 1.15, 95% CI 0.88 to 95% CI 1.64-2.20), as well as in high total health spending 1.52), nonagenarian (aOR 1.27, 95% CI 0.81 to 2.02) patients. settings. SHI, health insurance system; THS, tax-­based There was a higher mortality in SHI in octogenarian (aOR healthcare system. 1.29, 95% CI 1.10 to 1.59) and female (aOR 1.32, 95% CI 1.02 to 1.70) patients and in patients staying below 72 hours (aOR economic data do not vary within a given cluster, we chose 1.41, 95% CI 1.08 to 1.85). generalised estimation equations (GEE) with robust stan- dard errors to produce population average ORs. The survival analysis/GEE-­ based analyses were conducted using only robust estimators of the standard errors and not in the sense of robustness against violations of normality assumptions as for the robust methods (eg, Mann–Whitney tests) used for the univariate analyses. Model-1 includes only the ICU as panel. Model-2 includes patient specific factors (sex, age per year, SOFA Score per point, Frailty Score per CFS point and the admission diagnosis). Model-3 adds the health economic data (GDP per capita, HDI, health spending per capita and ICU beds per capita). Model-4 adds the treatment strategies (treatment limitations and organ support) and calculates only mortality. Adjusted aOR and respective 95% CI were obtained. Sensitivity analyses stratifying 30-­day mortality for patient-­ specific characteristics (figure 1), treatment strat- egies (figure 2) and health economic data (figure 3) were done. Risk model calibration accuracy/goodness of fit was evaluated graphically by stratification of patients into SOFA quartiles and comparison of observed versus expected events Figure 2 Sensitivity analyses stratifying 30-­day mortality for within risk strata (figure 4). All tests were two sided, and a p treatment strategies using generalised estimation equations value of <0.05 was considered statistically significant. Stata/IC producing population average ORs. The depicted aORs from V.16.1 was used for all statistical analyses. model-1 include only the intensive care unit as panel. The 30-­ day mortality was similar in patients with renal replacement therapy (RRT), without vasoactive drugs, intubated and non-­ RESULTS intubated paitents and patients on non-­invasive ventilation (NIV). There was a trend towards higher 30-­day mortality in Baseline risk distribution in SHI versus THS patients without RRT (aOR 1.29, 95% CI 1.04 to 1.60) and In total, 7817 patients from 16 European countries without NIV (aOR 1.50, 95% CI 1.16 to 1.94). SHI, health were included in this study. The relative frequencies of insurance system; THS, tax-­based healthcare system. the admission diagnoses varied between THS and SHI Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 3 Open access the SHI in all models (table 2; model-3 aOR 1.89, 95% CI 1.27 to 2.82; p=0.002). The crude rates for withholding treatment (34% vs 26%) were higher in SHI systems. This finding persisted in model-1 (aOR 1.44, 95% CI 1.09 to 1.90; p=0.01) but not after adjustment for patient-­specific variable in model-2 and after adjustment for the health economic data in model-3 (aOR 1.10, 95% CI 0.78 to 1.53; p=0.59; table 2). Mortality analysis The overall ICU mortality was 27% and the 30-­ day mortality was 41%. The ICU mortality was consistently higher in the SHI systems, after adjustment for patient-­ specific variables in model-2, health economic data in model-3 and in model-4 integrating patient-­specific data, health economic data as well as treatment strategies (aOR 1.50 95% CI 1.09 to 2.06; p=0.01). However, the 30-­day mortality was similar between THS and SHI in all models (table 2). Sensitivity analyses In sensitivity analyses, stratifying 30-­ day mortality for patient-­specific characteristics (figure 1), treatment strat- egies (figure 2) we observed no differences in 30-­day mortality between THS and SHI, except for a trend towards higher mortality (aOR 1.41, 95% CI 1.08 to 1.85) in patients staying below 72 hours in the ICU in SHI. However, in the sensitivity analyses stratifying for the health economic data (figure 3), the 30-­day mortality was higher in SHI, in settings with a low number of ICU beds Figure 4 Risk model calibration accuracy/goodness of fit per capita (aOR 2.17, 95% CI 1.57 to 3.02), low GDP per was evaluated graphically by stratification of patients into capita (aOR 2.17, 95% CI 1.42 to 3.58) and low HDI (aOR Sequential Organ Failure Assessment (SOFA) quartiles (Q1: 1.22, 95% CI 1.64 to 2.20), as well as in high total health SOFA 0–4; Q2:>4–7; Q3>7–10, Q4>10) and comparison of spending settings. observed versus expected events within risk strata. In the In the graphical calibration analysis both THS and SHI graphical calibration analysis, both tax-­based healthcare showed agreement for predicted ICU mortality proba- system (THS) and health insurance system (SHI) showed bility and observed events in SOFA risk quartiles, both in agreement of predicted intensive care unit (ICU) mortality probability and observed events in SOFA risk quartiles, all patients (figure 4A) and in patients without any treat- both in all patients (A) and in patients without any treatment ment limitation (figure 4B). limitation (B). systems (table 1). There were higher rates of nonagenar- DISCUSSION ians (11% vs 9%; p=0.002) in SHI systems. The country The aim of our study was to investigate possible differ- specific differences within the systems are descriptively ences in the characteristics, management and outcome shown in online supplemental tables 1 and 2. of VIPs between THS and SHI healthcare systems. In The rates of frail patients (CFS>4; 46% vs 41%; p<0.001), general, critically ill patients admitted to ICUs in SHI were as well as patients with disabilities (ADL <5; 30% vs 24%; older, frailer and more frequently suffered from cogni- p<0.001)) and cognitive decline (IQCODE≥3.5; 37% vs tive impairment. All the health economic data, including 29%; p<0.001), were higher in the SHI compared with the health spending per capita and the ICU beds avail- THS. The length of ICU stay was higher (90±162 vs able per capita, we evaluated were higher in THS. We 72±134 hours; p<0.001) in the SHI (table 1). were able to detect trends towards a more liberal use of organ support, especially RRT, in SHI. ICU mortality was Organ support and management in SHI versus THS higher in SHI even after adjustment for confounders, The rates of NIV use were similar in model-1 and model-3 while 30-­day mortality showed no difference. There was and showed a trend towards higher odds in the THS a relatively consistent trend towards excess mortality in (table 2). The rates of RRT were consistently higher in poorer SHI systems, which could be interpreted as there 4 Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 Open access Table 1 Baseline characteristics of the cohort n=4941 n=2876 Female sex n (%) 2282 (46) 1418 (49) 0.008 Age (years) Median (±IQR) 84 (5) 84 (5) <0.001 Age >90 n (%) 450 (9) 324 (11) 0.002 Frailty Score—CFS Median (±IQR) 4 (3) 4 (3) <0.001 Frailty (CFS>4) n (%) 2008 (41) 1317 (46) <0.001 ADL* Median (±IQR) 6 (1) 6 (2) 0.03 Disablitiy (ADL≤4) 462 (24) 446 (30) <0.001 IQCODE* Median (±IQR) 3.19 (0.61) 3.25 (0.75) <0.001 Cognitive decline (IQCODE ≥3.5) n (%) 460 (29) 493 (37) <0.001 SOFA Score Median (±IQR) 7 (6) 7 (6) 0.006 SOFA quartiles 0.007 SOFA 0–4 n (%) 1514 (31) 844 (30) SOFA 5–7 n (%) 1277 (26) 747 (26) SOFA 8–10 n (%) 1090 (23) 608 (21) SOFA>10 n (%) 972 (20) 667 (23) ICU length of stay (hours) Median (±IQR) 72 (134) 90 (162) <0.001 ICU stay <72 hours n (%) 2476 (50) 1299 (45) <0.001 CPS* Median (±IQR) 10 (7) 11 (7) 0.10 Admission diagnosis Respiratory failure 1128 (23) 772 (27) <0.001 Circulatory failure 657 (13) 423 (15) 0.08 Combined circulatory and respiratory failure 535 (11) 352 (12) 0.06 Sepsis 693 (14) 354 (12) 0.03 Multitrauma without head injury 92 (2) 55 (2) 0.98 Multitrauma with head injury 87 (2) 58 (2) 0.42 Head injury 117 (2) 78 (3) 0.35 Intoxication 13 (<1) 23 (<1) 0.001 Cerebral injury (non-­traumatic) 286 (6) 212 (7) 0.006 Emergency surgery 694 (14) 239 (8) <0.001 Other 637 (13) 310 (11) 0.006 Economic data Health spending per capita (US$) Median (±IQR) 4653 (1626) 5376 (389) <0.001 HDI Median (±IQR) 0.92 (0.05) 0.89 (0.04) <0.001 GDP per capita (US$) Median (±IQR) 41 030 (18 295) 41 760 (8262) <0.001 ICU beds Median (±IQR) 6.6 (2.0) 11.6 (5.2) <0.001 *Only available in VIP2 study. ADL, activities of daily living; CFS, Clinical Frailty Scale; CPS, Co-­morbidity and Polypharmacy Score; GDP, gross domestic product; HDI, human development index; ICU, intensive care unit; IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly; SHI, social heath insurance system; SOFA, Sequential Organ Failure Assessment; THS, tax-­based health service. Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 5 6 Open access Table 2 Generalised estimation equations producing population average ORs n (%) n (%) aOR (95% CI, p value) aOR (95% CI, p value) aOR (95% CI, p value) aOR (95% CI, p value) Outcome ICU mortality, n (%) 1247 (25) 849 (30) 1.52 (1.22 to 1.88;<0.001) 1.37 (1.08 to 1.72; 0.009) 1.39 (1.06 to 1.83; 0.02) 1.50 (1.09 to 2.06; 0.01) 30-­day mortality, n (%) 1917 (41) 1132 (42) 1.31 (1.06 to 1.62; 0.01) 1.16 (0.92 to 1.46; 0.22) 1.03 (0.77 to 1.38; 0.83) 0.89 (0.66 to 1.21; 0.47) Treatment limitation         Treatment withhold, n (%) 1265 (26) 975 (34) 1.44 (1.09 to 1.90; 0.01) 1.18 (0.80 to 1.72; 0.41) 1.10 (0.78 to 1.53; 0.59)   Treatment witdraw, n (%) 737 (15) 410 (14) 1.19 (0.92 to 1.54; 0.18) 1.04 (0.76 to 1.44; 0.79) 0.83 (0.56 to 1.22; 0.34)   Management strategies NIV, n (%) 1222 (25) 706 (25) 1.01 (0.77 to 1.32; 0.96) 1.00 (0.73 to 1.37; 0.997) 0.54 (0.36 to 0.80; 0.002) Mechanical ventilation, n (%) 2425 (49) 1562 (54) 1.37 (1.02 to 1.85; 0.04) 1.42 (1.03 to 1.95; 0.03) 2.06 (1.39 to 3.07;<0.001) Vasoactive drugs, n (%) 2900 (59) 1676 (58) 1.12 (0.86 to 1.44; 0.40) 1.25 (0.92 to 1.69; 0.15) 1.34 (0.89 to 2.02; 0.16) Renal replacement therapy, n (%) 460 (9) 373 (13) 1.63 (1.18 to 2.24; 0.003) 1.70 (1.22 to 2.39; 0.002) 1.89 (1.27 to 2.82; 0.002) Model-1 includes only the ICU as panel. Model-2 includes patient specific factors (sex, age per year, SOFA Score per point, Frailty Score per Clinical Frailty Scale point and the admission diagnosis). Model-3 adds the health economic data (GDP per capita, HDI and health spending per capita). Model-4 adds the treatment strategies (treatment limitations and organ support) and calculates only mortality. Adjusted OR (aOR) and respective 95% CI were obtained. Model 1—ICU cluster as random effect. Model 2—Model 1 plus patient level (sex, SOFA, age, frailty, admission diagnosis). Model 3—Model 2 plus economic data (health spending, HDI, GDP, ICU beds). Model 4—Model 3 plus management strategies (NIV, mechanical ventilation, renal replacement therapy, vasoactive drugs) and treatment limitation (treatment withdraw and/or withhold). GDP, gross domestic product; HDI, human development index; ICU, intensive care unit; NIV, non-­invasive ventilation; SOFA, Sequential Organ Failure Assessment. Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 Open access being an advantage in having a THS system in precarious ICU mortality in SHI systems as the result of a more settings where there are limited resources. liberal admissions policy due to the increased availability of ICU beds. This interpretation is supported by the fact Baseline risk distribution in SHI versus THS that in the ICU mortality analysis stratified according to We interpret our findings that patients admitted to ICUs the SOFA Score (figure 4), excess mortality in the SHI was in SHI were older, frailer and more frequently suffered primarily detected in the highest quartile (SOFA >10)— from cognitive impairment, as perhaps suggesting a more this quartile is also overrepresented in the SHI (23% vs liberal admission policy in SHI. Intensivists in THS might 20%). This could be due to a higher rate of ‘futile’ admis- need to apply a more stringent and strict triage criterion sions, which were a priori not associated with clinical prior to ICU admission, as SHI had a higher number of benefit. In other words, selection and triage, both by no ICU beds per capita. While in the THS, the allocation of means an easy task, in the THS may take place prior to resources is primarily centralised and, for example, based ICU admission and in SHI during the ICU admission.6 34 on quality-­adjusted life year for specific procedures, in the We are, however, aware, that triage is not only a clinical SHI there is a coexistence of public and private owner- but also an ethical challenge. Interestingly, the THS ship of hospitals.17 This gives rise to two possible explana- showed a trend towards higher rates of ‘short-­term’ ICU tions of the differences observed. First, the coexistence of stays (arbitrary cut-­off at 72 hours). Again, this finding public and private providers could lead to an excess (from can be interpreted in a number of ways. On the one the THS point of view) of poorly evidence-­based health- hand, this could primarily reflect the overall lower (and care interventions for economic reasons resulting in the thus more resource efficient) duration of ICU stays in the wrong incentives for patient care. Second, the focus on THS. On the other hand, it could reflect a higher rate of financial constraints and resulting therapeutic and equip- ‘ICU trials’, that is, admissions to ICUs with the rationale ment limitations in the THS (from the SHI point of view) to try a curative therapy trial for two to 2–3 days. At this could lead to a restriction on admission and treatment, point, the prognosis is re-­evaluated, and if continuing the especially to subgroups of patients (such as the elderly). current treatment is deemed futile, there is a change in The verification of one of these two considerations is the goal of therapy and the patient is discharged from beyond the scope of this paper. ICUs to receive ongoing care on a normal ward.35 After careful adjustment for possible confounders, we Organ support and management in SHI versus THS did not observe differences in the 30-­day mortality. Why The rates of RRT were consistently higher in SHI, but this the better survival data at the time of ICU discharge in broader application of organ support did not translate the THS do not translate into better outcomes at 30 days into a superior 30-­day mortality compared with the more is beyond the scope of the present study. It could reflect restrictive approach in THS. In our opinion, this could be different care for older patients in the THS (for example, interpreted as an indication of a more resource-­efficient because fewer geriatric or rehabilitation medical approach to patient care in THS. That this, presumably resources are available). If one accepted our consider- more economically favourable strategy, resulted in no ations from above regarding triage, the THS may ulti- detriment to patient outcome is interesting and poten- mately fail to translate the initially better patient selection tially thesis generating. into better long-­term outcomes. This could underline the The decision to restrict treatment based on a predicted importance of geriatric treatment, especially after survival individual prognosis is not an easy clinical task, even of an acute intensive care event. Intensive care certainly in older, critically ill patients.32 33 We observed a trend contributes to survival of the acute event, but only towards more therapy limitation in SHI systems in the specialist geriatric treatment and rehabilitation might be unadjusted analysis, but this signal disappeared after able to translate this into better long-­term outcomes. We adjustment for health economic data. This signal could can only speculate at this point about the morbidity rate, also have been interpreted as an excess of ICU beds in which is likely to also benefit from specialist geriatric care SHI. If one assumes that people in THS and SHI systems and rehabilitation.36 are of the same age with similar severity of illness, an The pronounced increase in mortality from 27% on increase in treatment goal limitations in SHI systems ICU discharge to 41% at 30 days could reflect the high could indicate an overly liberal allocation of intensive well documented known mortality of older, critically ill care resources, especially when the outcomes are similar. patients.2 3 Systematic factors such, as national healthcare It could then be assumed that patients with limitations frameworks, or even patient-­specific treatment strategies on ceilings of care in the THS are not primarily treated and therapy concepts could become less relevant in a in intensive care (and thus in a resource-­saving manner). group of very old patients as the age-­associated mortality inherent in humans and the observed mortality could Mortality analysis simply be an expression of the natural dying process. The trend we have observed towards ICU excess mortality in SHI would also fit this explanation but could also be Sensitivity analyses interpreted as an indication of worse quality of care in In the sensitivity analysis, there was a signal that patients SHI. We, however, interpret this trend towards higher with a shorter length of stay had a higher mortality in SHI Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 7 Open access ICUs. Therefore, although in SHI hospitals more patients Lastly, we are well aware of the limitations of sensitivity are admitted to ICU, they die relatively quickly. One and subgroup analyses, which limit the generalisability of possible explanation for this is that in THS ICUs these the conclusions drawn (figures 1–3). We chose to report patients are never admitted, as when triaged for admis- these results to help generate ideas for future public sion to ICU they are deemed unsuitable and therefore health analyses. Despite all the limitations we consider palliated on the wards. our analysis to be a solid foundation for further studies There was a trend towards higher 30-­day mortality, in comparing ICU management and its outcomes in both the sensitivity analyses, in countries with lower GDP, lower the THS and SHI healthcare settings. HDI and lower numbers of ICU beds available in the SHI. The relatively consistent trend towards excess mortality Conclusion in lower GDP SHI could be an expression of the relative To conclude, the adjusted 30-­day mortality was similar in superiority of a THS in settings with limited resources. both systems. Patients in SHI were older, sicker and frailer Future public health studies could test this hypothesis, as which could be interpreted as a sign of a more liberal it has potential consequences, for example, in assisting admission policy in SHI. We believe that the observed developing countries (with limited resources) that are trend towards ICU excess mortality in SHI results mainly building a new health system. from a more liberal admission policy and an increased use of treatment limitations. Limitations Author affiliations This retrospective, post hoc study has several limitations. 1 Department of Anaesthesiology, Perioperative Medicine and Intensive Care First, we are aware that the division between THS and Medicine, Paracelsus Medical University Salzburg, Salzburg, Austria SHI is an over-­ simplification and that mixed systems 2 Center for Public Health and Healthcare Research, Paracelsus Medical University exist, especially in some countries such as Austria and Salzburg, Salzburg, Austria 3 Department of Medical Intensive Care, Hadassah University Medical Center, Portugal. Further, we categorised the countries into SHI Jerusalem, Israel versus THS, but there may be country-­specific and even 4 Division of Cardiology, Pulmonology, and Vascular Medicine, University Hospital site-­ specific differences within the two health systems. Dusseldorf, Dusseldorf, Germany 5 We have shown these differences for the primary and General Intensive Care, Hadassah Medical Center, Jerusalem, Israel 6 secondary endpoints in the online supplemental tables. service de réanimation médicale, Hôpital Saint-­Antoine, Paris, France 7 Depatment of Intensive Care Medicine, Autonomous University of Barcelona, These differences are presented in a descriptive manner Sabadell, Catalonia, Spain as the absolute numbers in the countries are relatively 8 Department of Anaesthesia, Humanitas University, Milan, Italy low and not suitable for a formal statistical analysis. Other 9 Critical Care Medicine, Mater Misericordiae University Hospital, Dublin, Ireland 10 national or local factors that could influence our results Department of Intensive Care, UCINC, Hospital de São José, Centro Hospitalar de Lisboa Central, EPE, Lisbon, Portugal are religious/cultural differences and expectations of 11 Department of Intensive Care, University of Ghent, Ghent, Belgium family members. Unfortunately, we do not have any data 12 Department of Acute Medicine, Geneva University Hospitals, Geneve, Switzerland on these potential confounders. Ultimately, we believe 13 Department of Intensive Care and Perioperative Medicine, Jagiellonian University that the division into primarily directly tax-­ financed Medical College, Krakow, Poland 14 systems with strong state control (THS) or indirectly General Intensive Care, Research Lead Critical Care Directorate St George’s Hospital, London, UK financed social insurance systems (SHI) is justified. 15 Department of Medicine and Health Sciences, Faculty of Health Sciences, Another limitation is the potential selection bias. The Linköping University, Division of Cardiovascular Medicine, Linköping, Sweden 30-­day endpoint was not available for all patients, which 16 Department of Cardiothoracic Anaesthesia and Intensive Care, University Hospital, may have led to a selection bias. Furthermore, we have no Linköping, Sweden 17 specific information on local, national or inbuilt admis- Intensive Care Medicine, Inselspital, Bern University Hospital, Bern, Switzerland 18 Division o Intensive Care and Emergency Medicine, Dept. Medicine, Medizinische sion algorithms. For example, patients treated in SHI Universitat Innsbruck, Innsbruck, Tirol, Austria systems on ICUs (and triaged in the course of an ICU 19 Department of Intensive Care, Aarhus Universitet, Aarhus, Denmark trial) may well be treated in THS intermediate care units 20 Department of Critical Care, General University Hospital of Larissa, Larissa, not primarily managed by intensive care physicians and Thessaly, Greece 21 Department of Intensive Care Medicine, Utrecht University, Utrecht, The thus not included in the study data, which could also lead Netherlands to selection bias. 22 Service de Réanimation, Service de Réanimation Médicale Hôpital Saint-­Antoine, Furthermore, in some areas our data lacks granularity. Paris, France 23 There are no data on functional capacity and morbidity, Department of Anesthesia and Intensive Care, Haukeland University Hospital, or management (discharge to home or institution), and/ Bergen, Norway or staffing ratios. Also, no data are available on the char- Collaborators VIP2 study group: Michael Joannidis, Philipp Eller, Raimund Helbok, acteristics or absolute or relative frequencies of patients René Schmutz, Joke Nollet, Nikolaas de Neve, Pieter De Buysscher, Sandra Oeyen, refused ICU admission. Despite this, we do have data on Walter Swinnen, Marijana Mikačić, Anders Bastiansen, Andreas Husted, Bård the baseline frailty and functional capacity for a signifi- E S Dahle, Christine Cramer, Christoffer Sølling, Dorthe Ørsnes, Jakob Edelberg Thomsen, Jonas Juul Pedersen, Mathilde Hummelmose Enevoldsen, Thomas cant number of critically ill patients and we think that our Elkmann, Agnieszka Kubisz-­Pudelko, Alan Pope, Amy Collins, Ashok S Raj, Carole results add a novel insight into the differences between Boulanger, Christian Frey, Ciaran Hart, Clare Bolger, Dominic Spray, Georgina public healthcare systems. Randell, Helder Filipe, Ingeborg D Welters, Irina Grecu, Jane Evans, Jason Cupitt, 8 Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 Open access Jenny Lord, Jeremy Henning, Joanne Jones, Jonathan Ball, Julie North, Kiran Open access This is an open access article distributed in accordance with the Salaunkey, Laura Ortiz-­Ruiz De Gordoa, Louise Bell, Madhu Balasubramaniam, Creative Commons Attribution Non Commercial (CC BY-­NC 4.0) license, which Marcela Vizcaychipi, Maria Faulkner, McDonald Mupudzi, Megan Lea-­Hagerty, permits others to distribute, remix, adapt, build upon this work non-­commercially, Michael Reay, Michael Spivey, Nicholas Love, Nick Spittle Nick Spittle, Nigel and license their derivative works on different terms, provided the original work is White, Patricia Williams, Patrick Morgan, Phillipa Wakefield, Rachel Savine, Reni properly cited, appropriate credit is given, any changes made indicated, and the use Jacob, Richard Innes, Ritoo Kapoor, Sally Humphreys, Steve Rose, Susan Dowling, is non-­commercial. See: http://​creativecommons.​org/​licenses/​by-​nc/​4.​0/. Susannah Leaver, Tarkeshwari Mane, Tom Lawton, Vongayi Ogbeide, Waqas Khaliq, Yolanda Baird, Antoine Romen, Arnaud Galbois, Bertrand Guidet, Christophe ORCID iDs Vinsonneau, Cyril Charron, Didier Thevenin, Emmanuel Guerot, Guillaume Besch, Bernhard Wernly http://​orcid.​org/​0000-​0003-​4024-​0220 Guillaume Savary, Hervé Mentec, Jean-­Luc Chagnon, Jean-­Philippe Rigaud, Jean-­ Raphael Romano Bruno http://​orcid.​org/​0000-​0003-​3776-​3530 Pierre Quenot, Jeremy Castaneray, Jérémy Rosman, Julien Maizel, Kelly Tiercelet, Malte Kelm http://​orcid.​org/​0000-​0001-​5834-​2757 Lucie Vettoretti, Maud Mousset Hovaere, Messika Messika, Michel Djibré, Nathalie Wojciech Szczeklik http://​orcid.​org/​0000-​0002-​1349-​1123 Rolin, Philippe Burtin, Pierre Garcon, Saad Nseir, Xavier Valette, Christian Rabe, Eberhard Barth, Henning Ebelt, Kristina Fuest, Marcus Franz, Michael Horacek, Michael Schuster, Patrick Meybohm, Raphael Romano Bruno, Sebastian Allgäuer, Simon Dubler, Stefan J Schaller, Stefan Schering, Stephan Steiner, Thorben Dieck, Tim Rahmel, Tobias Graf, Anastasia Koutsikou, Aristeidis Vakalos, Bogdan Raitsiou, REFERENCES Elli Niki Flioni, Evangelia Neou, Fotios Tsimpoukas, Georgios Papathanakos, Giorgos 1 Flaatten H, Garrouste-­Orgeas M. The very old ICU patient: a never-­ Marinakis, Ioannis Koutsodimitropoulos, Kounougeri Aikaterini, Nikoletta Rovina, ending story. Intensive Care Med 2015;41:1996–8. Stylliani Kourelea, Tasioudis Polychronis, Vasiiios Zidianakis, Vryza Konstantinia, Zoi 2 Boumendil A, Angus DC, Guitonneau A-­L, et al. Variability of Aidoni, Brian Marsh, Catherine Motherway, Chris Read, Ignacio Martin-­Loeches, intensive care admission decisions for the very elderly. PLoS One Andrea Neville Cracchiolo, Aristide Morigi, Italo Calamai, Stefania Brusa, Ahmed 2012;7:e34387. Elhadi, Ahmed Tarek, Ala Khaled, Hazem Ahmed, Wesal Ali Belkhair, Alexander D 3 Leblanc G, Boumendil A, Guidet B. Ten things to know about Cornet, Diederik Gommers, Dylan de Lange, Eva van Boven, Jasper Haringman, critically ill elderly patients. Intensive Care Med 2017;43:217–9. 4 Boumendil A, Somme D, Garrouste-­Orgeas M, et al. Should elderly Lenneke Haas, Lettie van den Berg, Oscar Hoiting, Peter de Jager, Rik T Gerritsen, patients be admitted to the intensive care unit? Intensive Care Med Tom Dormans, Willem Dieperink, Alena Breidablik Alena Breidablik, Anita Slapgard, 2007;33:1252. Anne-­Karin Rime, Bente Jannestad, Britt Sjøbøe, Eva Rice, Finn H Andersen, 5 Guidet B, de Lange DW, Flaatten H. Should this elderly patient be Hans Frank Strietzel, Jan Peter Jensen, Jørund Langørgen, Kirsti Tøien, Kristian admitted to the ICU? Intensive Care Med 2018;44:1926–8. Strand, Michael Hahn, Pål Klepstad, Aleksandra Biernacka, Anna Kluzik, Bartosz 6 van Heerden PV, Sviri S, Beil M, et al. The wave of very old people in Kudlinski, Dariusz Maciejewski, Dorota Studzińska, Hubert Hymczak, Jan Stefaniak, the intensive care unit-­A challenge in decision-­making. J Crit Care Joanna Solek-­Pastuszka, Joanna Zorska, Katarzyna Cwyl, Lukasz J Krzych, Maciej 2020;60:290–3. 7 Hollinger A, Gayat E, Féliot E, et al. Gender and survival of critically Zukowski, Małgorzata Lipińska-­Gediga, Marek Pietruszko, Mariusz Piechota, Marta ill patients: results from the FROG-­ICU study. Ann Intensive Care Serwa, Miroslaw Czuczwar, Mirosław Ziętkiewicz, Natalia Kozera, Paweł Nasiłowski, 2019;9:43. Paweł Sendur, Paweł Zatorski, Piotr Galkin, Ryszard Gawda, Urszula Kościuczuk, 8 Block L, Petzold M, Syrous AN, et al. Age, saps 3 and female sex Waldemar Cyrankiewicz, Wojciech Gola, Alexandre Fernandes Pinto, Ana Margarida are associated with decisions to withdraw or withhold intensive care. Fernandes, Ana Rita Santos, Cristina Sousa, Inês Barros, Isabel Amorim Ferreira, Acta Anaesthesiol Scand 2019;63:1210–5. Jacobo Bacariza Blanco, João Teles Carvalho, Jose Maia, Nuno Candeias, Nuno 9 Jung C, Wernly B, Muessig JM, et al. A comparison of very old Catorze, Vladislav Belskiy, Africa Lores, Angela Prado Mira, Catia Cilloniz, David patients admitted to intensive care unit after acute versus elective Perez-­Torres, Emilio Maseda, Enver Rodriguez, Estefania Prol-­Silva, Gaspar Eixarch, surgery or intervention. J Crit Care 2019;52:141–8. 10 Ferrante LE, Pisani MA, Murphy TE, et al. Functional trajectories Gemma Gomà, Gerardo Aguilar, Gonzalo Navarro Velasco, Marián Irazábal Jaimes, among older persons before and after critical illness. JAMA Intern Mercedes Ibarz Villamayor, Noemí Llamas Fernández, Patricia Jimeno Cubero, Sonia Med 2015;175:523–9. López-­Cuenca, Teresa Tomasa, Anders Sjöqvist, Camilla Brorsson, Fredrik Schiöler, 11 Muessig JM, Nia AM, Masyuk M, et al. Clinical frailty scale (CFS) Henrik Westberg, Jessica Nauska, Joakim Sivik, Johan Berkius, Karin Kleiven reliably stratifies octogenarians in German ICUs: a multicentre Thiringer, Lina De Geer, Sten Walther, Filippo Boroli, Joerg C Schefold, Leila Hergafi, prospective cohort study. BMC Geriatr 2018;18:162. Philippe Eckert, Ismail Yıldız, Ihor Yovenko, Yuriy Nalapko, Yuriy Nalapko, Richard 12 Bruno RR, Wernly B, Flaatten H, et al. The hospital frailty risk score is Pugh of limited value in intensive care unit patients. Crit Care 2019;23:239. 13 Canevelli M, Raganato R, Remiddi F, et al. Counting deficits or Contributors BW, CJ and MB conceived the presented idea. BW developed the diseases? the agreement between frailty and multimorbidity theory and performed the computations. RRB, SB, MK, SS, PVvH, AB, AA, MC, BM, in subjects with cognitive disturbances. Aging Clin Exp Res RM, SO, BBP, WS, SL, SMW, JCS, MJ, JF, TZ, DdL, BG, HF and CJ contributed to 2020;32:179–82. the final version of the manuscript. CJ supervised the project. SL edited English 14 Flaatten H, De Lange DW, Morandi A, et al. The impact of frailty on language. CJ, HF and MB verified the analytical methods. ICU and 30-­day mortality and the level of care in very elderly patients (≥ 80 years). Intensive Care Med 2017;43:1820–8. Funding The authors have not declared a specific grant for this research from any 15 Delnoij DMJ. Bismarck or Beveridge: primary care matters. Eur J funding agency in the public, commercial or not-­for-­profit sectors. Public Health 2013;23:349. 16 Gaeta M, Campanella F, Capasso L, et al. An overview of different Competing interests None declared. health indicators used in the European health systems. J Prev Med Patient consent for publication Not required. Hyg 2017;58:E114–20. 17 van der Zee J, Kroneman MW. Bismarck or Beveridge: a beauty Ethics approval Each study site obtained institutional research ethics board contest between dinosaurs. BMC Health Serv Res 2007;7:94. approval. 18 Herzlinger RE, Parsa-­Parsi R. Consumer-­Driven health care: lessons from Switzerland. JAMA 2004;292:1213–20. Provenance and peer review Not commissioned; externally peer reviewed. 19 Health care systems in the European Union countries. Available: Data availability statement Data are available upon reasonable request. Data are https://www.​mscbs.​gob.​es/​estadEstudios/​estadisticas/​docs/​ presentacion_​en.​pdf available upon reasonable request. 20 Busse R, Schreyögg J, Gericke C. Analyzing changes in health Supplemental material This content has been supplied by the author(s). It has financing arrangements in high-­income countries a comprehensive not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been framework approach. 1. Articles in other Journals and Books, 2007. peer-­reviewed. Any opinions or recommendations discussed are solely those 21 Kulesher RR, Elizabeth Forrestal E, Forrestal E. International models of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and of health systems financing. J Hosp Adm 2014;3:127–39. 22 Wagstaff A. Social health insurance vs. tax-­financed health systems-­ responsibility arising from any reliance placed on the content. Where the content evidence from the OECD: the world bank, 2009. includes any translated material, BMJ does not warrant the accuracy and reliability 23 Joël M-­E, Dufour-­Kippelen S. Financing systems of care for older of the translations (including but not limited to local regulations, clinical guidelines, persons in Europe. Aging Clin Exp Res 2002;14:293–9. terminology, drug names and drug dosages), and is not responsible for any error 24 Guidet B, de Lange DW, Boumendil A, et al. The contribution of and/or omissions arising from translation and adaptation or otherwise. frailty, cognition, activity of daily life and comorbidities on outcome in Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909 9 Open access acutely admitted patients over 80 years in European ICUs: the VIP2 32 Guidet B, Flaatten H, Boumendil A, et al. Withholding or study. Intensive Care Med 2020;46:57–69. withdrawing of life-­sustaining therapy in older adults (≥ 80 years) 25 Rockwood K, Song X, MacKnight C, et al. A global clinical measure admitted to the intensive care unit. Intensive Care Med of fitness and frailty in elderly people. CMAJ 2005;173:489–95. 2018;44:1027–38. 26 Jorm AF, Jacomb PA. The informant questionnaire on cognitive 33 Nih consensus conference. diagnosis and treatment of depression in decline in the elderly (IQCODE): socio-­demographic correlates, late life. JAMA 1992;268:1018–24. reliability, validity and some norms. Psychol Med 1989;19:1015–22. 34 Flaatten H, Van Heerden V, Jung C, et al. The good, the bad and the 27 Katz S. Assessing self-­maintenance: activities of daily living, ugly: pandemic priority decisions and triage. J Med Ethics 2020. mobility, and instrumental activities of daily living. J Am Geriatr Soc doi:10.1136/medethics-2020-106489. [Epub ahead of print: 10 Jun 1983;31:721–7. 28 International-­Monetary-­Fund. World economic outlook reports, 2019. 2020] (published Online First: 2020/06/12). Available: https://www.​imf.​org/​en/​publications/​weo 35 Lecuyer L, Chevret S, Thiery G, et al. The ICU trial: a new admission 29 UNDP. Human development report 2019, 2019. policy for cancer patients requiring mechanical ventilation. Crit Care 30 OECD. Health at a glance, 2019. Available: https://​data.​oecd.​org/​ Med 2007;35:808–14. healthres/​health-​spending.​htm 36 Cuthbertson BH, Goddard SL, Lorencatto F, et al. Barriers and 31 Rhodes A, Ferdinande P, Flaatten H, et al. The variability of critical facilitators to early rehabilitation in the ICU: a theory driven Delphi care bed numbers in Europe. Intensive Care Med 2012;38:1647–53. study. Crit Care Med 2020;48:e1171–8. 10 Wernly B, et al. BMJ Open 2021;11:e046909. doi:10.1136/bmjopen-2020-046909