Background
Anaemia is a highly prevalent issue with deleterious impacts among critically ill patients [1, 2]. It is estimated that approximately 70% of patients who were admitted to the intensive care unit (ICU) experienced anaemia, which may result from phlebotomy, occult gastrointestinal bleeding, erythropoietin deficiencies, low iron levels, and coagulopathy [1, 3, 4]. Accumulating evidence have shown that anaemia in critically ill patients was associated with deleterious short-term adverse outcomes, including prolonged hospital length of stay, ventilator day, and increased hospital mortality [2, 5, 6]. Notably, anaemia appears to be a lasting issue given the current critical care strategy with restrictive red blood cell transfusion [4, 7, 8]. Recent longitudinal studies have further shown that anaemia may be persistent for nearly one year after critical illness [9, 10]. A number of studies have implicated anaemia with persistent inflammation and catabolism in patients who survived from critical illness, so-called chronic critical illness or post-intensive care syndrome [11, 12]. There is a research niche to explore the association between anaemia and long-term mortality in critically ill patients. Notably, the association between anaemia and mortality might be difficult to be clarified given the existence of a number of critical care relevant confounders in real-world data [13, 14]. In this study, we used both the critical care database at Taichung Veterans General Hospital (TCVGH) as well as the nationwide death-registry file in Taiwan to investigate the prevalence of anaemia within the first week and to explore the association between anaemia and long-term mortality in critically ill patients using propensity score-based statistical analyses.
Methods
Ethical approval
This study was approved by the Institutional Review Board at the Taichung Veterans General Hospital (SE20249B-1), and informed consent was waived, given that all of the data were de-identified data.
Anzeige
Patient population and definition of anaemia
This retrospective cohort study enrolled consecutive critically ill patients at TCVGH, a referral centre with approximately 1,600 beds in central Taiwan, between 2015 and 2020. Among patients with more than one ICU admission, the first ICU admission was used as the index ICU admission. Subjects without data on haemoglobin level in the first week of ICU admission were excluded from analyses. The presence of anaemia was defined by the mean level of week-1 haemoglobin lower and higher/equal than 10 g/dL given that 7–9 g/dL is currently recommended target level of haemoglobin in critical care [4, 8, 9, 15].
Primary outcome
The outcome of interest in the present study was the time to all-cause mortality. The date of death was obtained from the death registration profile of the National Health Insurance Database (NHID) in Taiwan [16]. In brief, Taiwan has implemented a compulsory National Health Insurance (NHI) program since 1995 with nearly total coverage of the population in Taiwan; therefore, we think the date of death in this study should be highly accurate.
Definition of the covariate
We used the critical care database at TCVGH to retrieve critical data relevant variables, consisting of demographic data, laboratory data, comorbidities in accordance with the Charlson Comorbidity Index (CCI) [17], Acute Physiology and Chronic Health Evaluation (APACHE) II score, presence of shock defined by the use of vasopressor, days of receiving mechanical ventilation, and managements including blood transfusion during admission.
Statistical analyses
The continuous data were presented as means ± standard deviation, and the categorical data were presented as numbers (percentages). Kaplan-Meier curves in patients whose week-1 haemoglobin was higher or lower than 10 g/dL were plotted. A Cox proportional hazards model was constructed after testing the proportional hazard assumption based on the Schoenfeld residuals [18]. We then determined hazard ratios (HRs) and 95% confidence intervals (CIs) for the overall mortality after adjusting potential confounders, including age, sex, comorbidities, and the other potential cofounders. Statistical analyses were conducted by R (3.6.0), and the level of significance was 0.05.
Anzeige
Propensity score-based analyses
We used propensity score matching (PSM) and weighting methods, consisting of inverse probability of treatment weight (IPTW) as well as covariate balancing propensity score (CBPS) to explore the association between week-1 anaemia and the long-term overall mortality among enrolled patients [19, 20]. The optimal nearest neighbour matching algorithm was used in PSM, and the calliper distance of the standard mean difference was set at 0.20. In brief, PSM was designed for nonexperimental causal studies by reducing the multidimensional covariate space for the probability of week-1 anaemia or not in the present study into one dimension. However, a number of cases might be excluded from analyses due to the lack of matched control subjects, and the restricted sub-population might raise the concern that the selected population might be distinct from the intended population [19]. Therefore, propensity score weighting methods, including IPTW, have been proposed to enrol the whole population; however, the high weight at the tails of the propensity score distribution could reduce the balance among covariates [21, 22]. Recently, CBPS is increasingly applied given that CBPS enrol the whole population through weighting as well as optimising the covariate balance [20, 23].
Subgroup analysis
We employed the Wald test to measure the significance of the modification effect of variables, including age, sex, renal disease, liver disease, pulmonary disease, the severity of critical illness, blood transfusion, and haemoglobin-associated laboratory data.
Results
Characteristics of the enrolled critically ill patients and the propensity score-matched population
Figure 1 illustrates the subject enrollment process of the original population (n = 7,089) and propensity core-matched population (n = 3,474) (Fig. 1). A total of 7,089 patients were eligible for analyses, and 45.0% (3,189/7,089) of them had anaemia. Patients with anaemia were older (67.5 ± 16.1 vs. 64.6 ± 15.9 years), had lower body mass index (BMI) (23.8 ± 4.6 vs. 24.6 ± 4.7), were less likely to be male (58.9% vs. 68.1%), had more comorbidities compared those without anaemia. The critical illness-associated variables, including APACHE II score, presence of shock, presence of sepsis, use of mechanical ventilation, leukocytosis, thrombocytopenia, low level of albumin and elevated level of blood urea nitrogen, were more severe in patients with anaemia than those without anaemia. After the PSM, most of variables were comparable between the two groups, although the anaemia group remained has a slightly increased proportion of renal diseases, haematological malignancy, and liver diseases (Table 1).
Fig. 1
Flowchart of subject enrollment Abbreviations: ICUs, intensive care units; TCVGH, Taichung Veterans General Hospital; Hb, Haemoglobin (g/dL)
×
![]()
Table 1
Characteristics between the patients categorised by week-1 level of haemoglobin in the primary cohort and propensity score-matched cohort
Before PSM | 1:1 PSM | ||||||
|---|---|---|---|---|---|---|---|
Hb ≥ 10 | Hb < 10 | SMD | Hb ≥ 10 | Hb < 10 | SMD | ||
n = 3,900 | n = 3,189 | n = 1,685 | n = 1,685 | ||||
Basic characteristics
| |||||||
Age, years | 64.6 ± 15.9 | 67.5 ± 16.1 | 0.186 | 68.1 ± 15.6 | 67.2 ± 16.5 | 0.058 | |
Sex (male) | 2656 (68.1%) | 1879 (58.9%) | 0.192 | 1085 (62.5%) | 1085 (62.5%) | < 0.001 | |
Body mass index | 24.6 ± 4.7 | 23.4 ± 4.6 | 0.257 | 23.4 ± 4.6 | 23.5 ± 4.6 | 0.020 | |
Comorbidities
| |||||||
Diabetes mellitus | 1200 (30.8%) | 1148 (36%) | 0.111 | 603 (34.7%) | 613 (35.3%) | 0.012 | |
Myocardial infarction | 1283 (32.9%) | 481 (15.1%) | 0.427 | 313 (18%) | 322 (18.5%) | 0.013 | |
Congestive heart failure | 599 (15.4%) | 521 (16.3%) | 0.027 | 309 (17.8%) | 284 (16.4%) | 0.038 | |
Valvular heart disease | 168 (4.3%) | 125 (3.9%) | 0.020 | 4 (0.5%) | 4 (0.5%) | < 0.001 | |
Dementia | 180 (4.6%) | 191 (6%) | 0.061 | 122 (7%) | 109 (6.3%) | 0.030 | |
Cerebrovascular disease | 923 (23.7%) | 563 (17.7%) | 0.149 | 356 (20.5%) | 341 (19.6%) | 0.022 | |
Hemiplegia | 114 (2.9%) | 51 (1.6%) | 0.089 | 38 (2.2%) | 36 (2.1%) | 0.008 | |
Peripheral vascular disease | 230 (5.9%) | 194 (6.1%) | 0.008 | 118 (6.8%) | 118 (6.8%) | < 0.001 | |
Renal disease | 562 (14.4%) | 1209 (37.9%) | 0.555 | 412 (23.7%) | 476 (27.4%) | 0.085 | |
COPD | 261 (6.7%) | 155 (4.9%) | 0.079 | 118 (6.8%) | 98 (5.6%) | 0.048 | |
Lymphoma/leukaemia | 80 (2.1%) | 304 (9.5%) | 0.324 | 74 (4.3%) | 113 (6.5%) | 0.100 | |
Metastatic solid tumour | 475 (12.2%) | 727 (22.8%) | 0.282 | 358 (20.6%) | 369 (21.2%) | 0.016 | |
Rheumatic disease | 126 (3.2%) | 229 (7.2%) | 0.179 | 91 (5.2%) | 94 (5.4%) | 0.008 | |
Peptic ulcer | 303 (7.8%) | 466 (14.6%) | 0.218 | 208 (12%) | 210 (12.1%) | 0.004 | |
Mild liver disease | 322 (8.3%) | 548 (17.2%) | 0.270 | 228 (13.1%) | 264 (15.2%) | 0.059 | |
Moderate or severe liver disease | 74 (1.9%) | 246 (7.7%) | 0.274 | 68 (3.9%) | 102 (5.9%) | 0.091 | |
Severity and managements
| |||||||
APACHE II score | 22.4 ± 5.6 | 26.3 ± 6.6 | 0.651 | 24.4 ± 6 | 24.8 ± 6 | 0.067 | |
Presence of sepsis | 371 (9.5%) | 694 (21.8%) | 0.342 | 345 (19.9%) | 345 (19.9%) | < 0.001 | |
Presence of shock | 1462 (37.5%) | 2055 (64.4%) | 0.560 | 1023 (58.9%) | 1023 (58.9%) | < 0.001 | |
Receiving mechanical ventilation | 1435 (36.8%) | 2042 (64%) | 0.566 | 1041 (59.9%) | 1039 (59.8%) | 0.002 | |
Receiving RBC transfusion | 638 (16.4%) | 2432 (76.3%) | 1.503 | 495 (28.5%) | 1213 (69.8%) | 0.908 | |
RBC transfusion (units) | 2.8 ± 1.1 | 2.4 ± 0.9 | 0.409 | 2.9 ± 1.1 | 2.5 ± 1.1 | 0.377 | |
Laboratory data
| |||||||
White blood cell count (103/µl) | 10.6 ± 5.3 | 12.6 ± 5.7 | 0.170 | 12.0 ± 6.3 | 12.2 ± 16.2 | 0.016 | |
Platelet (103/µL) | 210.9 ± 87.4 | 159.3 ± 108.9 | 0.523 | 189.2 ± 94.7 | 180.8 ± 117.4 | 0.079 | |
Albumin (mg/dL) | 3.6 ± 0.8 | 2.9 ± 0.7 | 1.063 | 3.1 ± 0.6 | 3.1 ± 0.7 | 0.045 | |
Blood urea nitrogen (mg/dL) | 24.5 ± 18.3 | 48 ± 33.6 | 0.870 | 33.1 ± 23.2 | 38.4 ± 29.8 | 0.195 | |
Creatinine (mg/dL) | 1.4 ± 1.6 | 2.9 ± 2.8 | 0.620 | 1.9 ± 2.2 | 2.3 ± 2.6 | 0.139 | |
One-year mortality
| 988 (25.3%) | 1867 (58.5%) | 0.715 | 732 (42.1%) | 906 (52.2%) | 0.202 | |
Association between anaemia and one-year mortality
Figure 2 illustrates the matching quality that was demonstrated by the measurement of the standardised mean difference (SMD) of variables between the two groups in the four distinct cohorts (Fig. 2). In the original population, several variables between patients with and without anaemia were apparently imbalanced. After matching as well as weighting, the aforementioned variables tended to be balanced in both PSM and IPTW, with the SMD lower than 0.20. The application of CBPS further reduced the imbalance of covariates between the two groups. We then focused on investigating the association between anaemia and one-year mortality among enrolled subjects in the aforementioned four populations. We confirmed the fulfilment of the proportional hazard assumption and then adjusted the covariates, including age, sex, and comorbidities in model 2, disease severity and management in model 3, and laboratory findings in model 4 (Supplemental Fig. 1 and Table 2). Table 2 summarises the consistent association between anaemia and risk for one-year mortality in the four populations. The adjusted HRs in the original PSM, IPTW and CBPS populations were 1.345 (95% CI 1.227–1.474), 1.265 (95% CI 1.145–1.397), 1.276 (95% CI 1.142–1.427) and 1.260 (95% CI 1.125–1.411), respectively.
Figure 2
The standardised mean difference between the patients whose week-1 haemoglobin were higher and lower than 10 g/dL in each cohort. Variables were ranked by the SMD. Abbreviations: SMD, standard mean difference; PSM, propensity score matching; IPTW, inverse probability of treatment weighting; CBPS, covariate balancing propensity score; APACHE, acute physiology and chronic health evaluation; BUN, blood urea nitrogen; BMI, body mass index; COPD, chronic obstructive pulmonary disease
×
![]()
Table 2
Cox proportional hazard regressions for estimation of the association between level of week-1 haemoglobin lower than 10 g/dL and one-year mortality in critically ill patients
Primary population | PSM population | IPTW population | CBPS populaiton | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
HR (95%CI) | p value | HR (95%CI) | p value | HR (95%CI) | p value | HR (95%CI) | p value | |||||
Model 1
| 2.950 (2.730–3.187) | < 0.001 | 1.337 (1.213–1.474) | < 0.001 | 1.241 (1.076–1.433) | 0.003 | 1.454 (1.304–1.622) | < 0.001 | ||||
Model 2
| 2.175 (1.998–2.368) | < 0.001 | 1.334 (1.209–1.471) | < 0.001 | 1.253 (1.103–1.424) | 0.001 | 1.356 (1.216–1.513) | < 0.001 | ||||
Model 3
| 1.706 (1.564–1.861) | < 0.001 | 1.308 (1.185–1.443) | < 0.001 | 1.271 (1.126–1.436) | < 0.001 | 1.318 (1.178–1.476) | < 0.001 | ||||
Model 4
| 1.345 (1.227–1.474) | < 0.001 | 1.265 (1.145–1.397) | < 0.001 | 1.276 (1.142–1.427) | < 0.001 | 1.260 (1.125–1.411) | < 0.001 | ||||
Anzeige
Analyses of interactive effects among patients in the PSM cohort
We further conducted subgroup analyses to explore the potential interaction effect of factors, particularly comorbidities that remained imbalanced in the PSM cohort and haemoglobin-associated variables (Table 3). We found that the majority of variables had no modification effect except for moderate/severe liver disease (p = 0.004). The association between anaemia within the first week and one-year mortality mainly existed in patients without liver disease (HR 1.361, 95% CI 1.230–1.506), whereas the association no longer exist in patients with moderate/severe liver disease (HR 0.806, 95% CI 0.560–1.159).
Table 3
Stratified analyses for modification effect on the association between level of week-1 haemoglobin and one-year mortality among the 3,440 subjects in the PSM cohort
Variables | Subgroup | Number of patients | Events (death) | HR ratio (95%CI) | p value | p for interaction |
|---|---|---|---|---|---|---|
Age (years) | 0.586 | |||||
< 65 | 1388 | 612 | 1.367 (1.164–1.605) | < 0.001 | ||
≥ 65 | 2086 | 1026 | 1.325 (1.172–1.498) | < 0.001 | ||
Sex
| 0.168 | |||||
Female | 1304 | 531 | 1.210 (1.020–1.435) | 0.028 | ||
Male | 2170 | 1107 | 1.415 (1.256–1.593) | < 0.001 | ||
Renal disease
| 0.588 | |||||
No | 2586 | 1268 | 1.328 (1.189–1.483) | < 0.001 | ||
Yes | 888 | 370 | 1.433 (1.163–1.765) | 0.001 | ||
COPD
| 0.167 | |||||
No | 3258 | 1531 | 1.361 (1.230–1.506) | < 0.001 | ||
Yes | 216 | 107 | 1.031 (0.705–1.508) | 0.877 | ||
Lymphoma/leukaemia
| 0.575 | |||||
No | 3287 | 1513 | 1.316 (1.189–1.456) | < 0.001 | ||
Yes | 187 | 125 | 1.420 (0.985–2.048) | 0.060 | ||
Moderate or severe liver disease
| 0.004 | |||||
No | 3304 | 1519 | 1.361 (1.230–1.506) | < 0.001 | ||
Yes | 170 | 119 | 0.806 (0.560–1.159) | 0.245 | ||
APACHE II score* | 0.765 | |||||
≤ 23.00 | 737 | 275 | 1.388 (1.095–1.760) | 0.007 | ||
> 23.00 | 2737 | 1363 | 1.315 (1.182–1.463) | < 0.001 | ||
Blood transfusion
| 0.248 | |||||
No | 1766 | 688 | 1.219 (1.039–1.429) | 0.015 | ||
Yes | 1708 | 950 | 1.082 (0.939–1.246) | 0.275 | ||
Platelet (103/µL)* | 0.128 | |||||
< 173.8 | 1737 | 990 | 1.359 (1.197–1.542) | < 0.001 | ||
≥ 173.8 | 1737 | 648 | 1.187 (1.017–1.384) | 0.029 | ||
Blood urea nitrogen (mg/dL)* | 0.799 | |||||
< 27.0 | 1707 | 691 | 1.326 (1.142–1.540) | < 0.001 | ||
≥ 27.0 | 1767 | 947 | 1.279 (1.124–1.455) | < 0.001 | ||
Creatinine (mg/dL)* | 0.699 | |||||
< 1.2 | 1660 | 700 | 1.289 (1.111–1.494) | 0.001 | ||
≥ 1.2 | 1814 | 938 | 1.305 (1.145–1.486) | < 0.001 |
Discussion
Anaemia is highly prevalent in critically ill patients, but the long-term mortality impact remains unclear. We used propensity score-based matching as well as weighting methods to address the association between anaemia within the first week and long-term mortality in critically ill patients. We identified that week-1 haemoglobin lower than 10 g/dL was associated with an increased risk of one-year mortality, and the association was robust in distinct propensity score-based methods consisting of PSM, IPTW and CBPS. Our findings point out the previously ignored long-term effect of anaemia and indicate that anaemia can be used for long-term mortality risk stratification in critically ill patients.
Anaemia is a prevalent and lasting issue in critically ill patients. Anaemia may result from complex critical illness-associated predisposing factors, including frequent phlebotomy, coagulopathy, blood loss, persistent/dysregulated inflammation, impaired erythropoietic response, kidney injury, and nutritional deficiencies [4, 24, 25]. Moreover, anaemia appears to be a lasting issue among critically ill patients, given that conservative red blood cell transfusion with a restrictive transfusion threshold (7 g/dL) is currently recommended based on the increasing evidence to show potentially detrimental effects of blood transfusion on the short-term outcome in critically ill patients [4, 8, 26]. In the present study, we found that 45.0% (3,189/7,089) of critically ill patients had anaemia, and the relatively high prevalence of anaemia might result from that we used haemoglobin level lower than 10 g/dL to define anaemia. Similarly, Warner et al. recently analysed haemoglobin levels not only during hospitalisation but also 12 months after discharge in 6,901 critically ill patients and found that the prevalence of mild anaemia, defined by haemoglobin level lower than 10 g/dL, at 3-month, 6-month and 12-month after discharge were 56%, 52% and 45%, respectively [9]. Moreover, Warner et al. found that rates of recovery from anaemia at 12 months in patients with mild (10.0–12 g/dL), moderate (8.0–10.0 g/dL) and severe anaemia (< 8.0 g/dL) were 58%, 39% and 24%, respectively [9]. These evidence suggest that anaemia is a lasting issue in critical care and highlight the need to study the long-term effect of anaemia in critically ill patients.
Anzeige
Recently, accumulating studies have used big real-world data to explore critical care-associated issues without large-scale randomised control trials [13, 14]. For example, critical care databases, including Medical Information Mart for Intensive Care (MIMIC) and eICU Collaborative Research Database (eICU-CRD), have been widely used to investigate crucial issues with respect to critical care [27, 28]. However, long-term mortality, a high-priority issue in critical care, cannot be assessed in the aforementioned critical care databases, given that the majority of open data are unidentifiable data that cannot be linked with the other databases [29, 30]. In the present study, the nationwide coverage of compulsory national health insurance in Taiwan enables us to link the critical care database at TCVGH with the Taiwanese NHID and to ascertain the long-term mortality among enrolled subjects. But, the existence of confounders is inevitable in real-world data, and we think it would be imprudent to draw a conclusion without the strenuous effort to mitigate the impact of confounders in real-world data [13, 31]. Collectively, real-world critical care databases are valuable for exploring crucial issues in critically ill patients, but additional efforts are required to mitigate the effect of confounders in real-world data.
In this study, we used three distinct propensity score-based methods to investigate the independent association between anaemia within the first week and one-year mortality in critically ill patients. PSM is a widely used PS-based method to obtain two comparable subgroups through matching; however, a highly stringent matching process in PSM may exclude the high number of enrolled subjects and might restrict the generalizability; therefore, we choose 0.20 as the set calliper distance of the standard mean difference in the present study [19]. Intriguingly, we noted that the strength of association between anaemia within the first week and one-year mortality in the PSM cohort tends to be lower than the strength of association in the primary cohort, IPTW cohort and CBPS (Table 2). We postulated that excluding nearly 50% of the original population during the process of matching might lead to a slight difference in the strength of association despite the similar trend of association (Supplemental Fig. 2). In addition to the matching method, we also employed weighting methods, including IPTW and CBPS, in this study to explore the independent association between anaemia within the first week and one-year mortality among the whole enrolled subjects with a critical illness [21, 22]. In IPTW, all of the subjects were included in analyses, and the PS was used as weights in the analyses to minimise the effects of observed confounding, although the imbalance of the covariate remains a concern, particularly variables with extreme weights at the tails of the propensity score distribution (Fig. 2) [21]. Therefore, we further used CBPS to mitigate the imbalance among covariates and found that the strength of association between anaemia within the first week and one-year mortality between IPTW and CBPS appear to be similar, indicating that the covariates between the two groups should be largely comparable in the present study [20]. Taken together, we used four statistical analyses to demonstrate a consistent association between week-1 anaemia and one-year mortality in critically ill patients, and the slight difference in the strength of association should reflect the distinct statistical approaches and assumptions.
Limitations
There are limitations in this study. First, this study is a single-centre study in central Taiwan, and more studies are warranted to validate our findings. Second, selection bias could be a concern due to the retrospective study design, but we have performed the propensity score-based matching and weighting analyses to guarantee the robustness of our results. Third, potential unmeasured confounders, such as alcohol intake and fluctuation of haemoglobin level, cannot be assessed; however, the adjustment for numerous variables and propensity score-based approach should at least partly mitigate this concern. Moreover, studies focusing on critically ill patients with moderate/severe liver diseases are warranted.
Future directions
Our findings shed light on the long-term relevance of anaemia and provide clinical evidence for risk stratification of the long-term mortality by week-1 variables in critically ill patients. More studies are warranted to clarify underlying biological mechanisms and to explore the potential effect of correction of anaemia within and after the first week of ICU admission.
Anzeige
Conclusions
In conclusion, anaemia is a substantial issue in critical illness, but the long-term effect on mortality of anaemia remains a research niche. In the present study, we linked the hospital-based critical care database at TCVGH with the Taiwanese nationwide death registry profile as well as propensity score-based methods to demonstrate that anaemia within the first week was associated with an increased risk of long-term mortality in critically ill patients.
Acknowledgements
Not applicable.
Declarations
Ethics approval and consent to participate
The study was performed in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of the Taichung Veterans General Hospital (SE20249B-1). All the data were anonymised data, and informed consent was waived by the Institutional Review Board of Taichung Veterans General Hospital.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests of the authors.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.