Evaluating compliance with restrictive blood transfusion practices: A 5‐year study from a Greek tertiary hospital

Abstract

Background and Objectives

Our aim was to retrospectively assess the adherence to a restrictive transfusion strategy (in the context of patient blood management [PBM]) among patients in a Greek tertiary hospital over a 5‐year period (2020–2024) following the initial evaluation of the electronic transfusion request system.

Materials and Methods

Analyses included assessment of time trends—overall and by hospital clinic—in haemoglobin (Hb) levels at transfusion, as well as in the proportion of inappropriate transfusions, based on pre‐specified audit criteria (Hb thresholds for red blood cell transfusions).

Results

A total of 23,322 transfusion episodes were recorded among medical patients, with annual counts ranging from 4078 to 5394 and an overall mean transfusion Hb level of 7.18 ± 0.78 g/dL. The proportion of transfusions with Hb levels above 7.00 g/dL decreased significantly from 69.14% in 2020 to 58.83% in 2023 (Cochran–Armitage test for trend, p < 0.01). Among patients without comorbidities (threshold: 7.00 g/dL), 53.34% of transfusions occurred above this cut‐off, with an initial drop from 60.41% in 2020 to 47.72% in 2021, followed by a rebound to 58.43% in 2024 (p = 0.25). Transfusions above 8.00 g/dL also declined significantly, from 8.33% in 2020 to 5.20% in 2023 (p < 0.01). Differences were also observed among clinics.

Conclusion

Overall, moderate compliance with the restrictive transfusion strategy was observed. The initial decline in transfusions above prespecified Hb thresholds followed by fluctuations in subsequent years highlights the need for a consistent, multifaceted effort to optimize blood usage.

Highlights.

• Electronic systems in hospitals can enhance transfusion decision making by integrating clinical judgement into the process.

• In a tertiary Greek hospital, compliance of clinics with a restrictive transfusion strategy was found to be moderate but has shown a declining trend in recent years.

• Ongoing effort is essential to maintain established milestones and ultimately to optimize blood usage.

INTRODUCTION

Transfusion strategies vary notably not only between comparable countries but also among hospitals within the same country for matched patients under the same medical conditions. An increasing volume of data demonstrates that transfusion is associated—in a dose‐dependent manner—with increased morbidity, mortality and hospital length of stay, with current evidence suggesting a causal association between unnecessary transfusions and worse clinical outcome [1]. The expanding ageing population in developed countries results in a growing gap between blood supply and demand, thereby placing greater strain on blood transfusion services. Furthermore, given the adverse transfusion outcomes, the weak evidence for beneficial transfusions and the high cost of blood products transfused, the need to optimize patient blood management (PBM) becomes apparent. Under these circumstances, focusing on a restrictive red blood cell (RBC) transfusion strategy, optimal blood transfusion has emerged as a critical issue for medical inpatients [2, 3, 4].

PBM is defined as ‘the timely application of evidence‐informed medical and surgical concepts designed to maintain haemoglobin (Hb) concentration, optimize haemostasis and minimize blood loss in an effort to improve patient outcomes’ [5]. Thus, PBM aims to shift focus from blood components to the patient, who assumes a central role [6]. PBM is based on three elements: optimization of patient's own blood production, minimization of blood loss, and optimization of the patient‐specific tolerance to anaemia, allowing strict adherence to physiological transfusion thresholds, thus minimizing exposure to transfusions [7].

PBM goals can be achieved only through a multiprofessional, multidisciplinary, patient‐centred strategy involving, besides transfusion medicine specialists, any medical staff participating in the diagnostic and therapeutic care aimed at PBM implementation. It is obvious that among the three elements composing PBM, only adherence to prespecified transfusion thresholds can be effectively controlled by the transfusion service staff. The patient's individualized features affecting transfusion decisions should also be considered, enabling a more ‘patient‐based’ approach. Considering the difficulties in collaborating with other medical specialists involved in PBM, the most efficient and immediately available method for the transfusion service to ensure optimal use of blood components is to evaluate the appropriateness of each transfusion. On the other hand, relying solely on an absolute Hb threshold level as the trigger for transfusion is not considered best practice.

Under these circumstances, 6 years ago (in 2019) the Transfusion Medicine Committee (TMC) of Attikon University Hospital in Athens, Greece, established local transfusion guidelines by adopting a restrictive RBC transfusion threshold. At the same time, the hospital's transfusion service implemented an electronic transfusion request form, enabling prospective auditing of transfusion practices. After a 1‐year implementation period, a retrospective review was performed, assessing the adopted transfusion practices regarding medical patients. Findings showed poor adherence to the restrictive transfusion threshold [8]. Considering the Hb level of 7.0 g/dL as the RBC transfusion trigger in medical non‐bleeding patients without comorbidities, inappropriate transfusions took place in about two‐thirds of patients. This proportion significantly decreased when the Hb level of 8.0 g/dL was used as the transfusion trigger, which is also a reasonable value adhering to a restrictive transfusion strategy in hospitalized, stable patients [9].

This analysis aims to retrospectively assess 5 years after the initial evaluation the adherence to the restrictive transfusion strategy adopted for medical patients using the electronic transfusion request system.

MATERIALS AND METHODS

This is a single‐centre retrospective study conducted at Attikon University Hospital, a tertiary hospital in Athens, the capital city of Greece. The Institutional Review Board of the hospital approved the study (6 May 2025, A6). The analysis examined the total number of RBC units transfused and the mean number of units transfused per case for a 5‐year period (2020–2024), after the initial evaluation of the electronic order system. In 2019, the hospital's blood transfusion service, in collaboration with the Hospital Information Technology Department, developed and implemented a programme to monitor and regulate transfusion practices. The TMC of the hospital established local transfusion guidelines and introduced an electronic transfusion request system for inpatients. A new computerized order form for transfusions was then introduced, requiring physicians to select at least one indication for each blood component ordered and to include individualized relevant laboratory and clinical information. Data for these orders were obtained from the laboratory information system or directly from clinical staff during the request process. A prospective audit was conducted, reviewing transfusion orders in real time before blood components were issued. If a transfusion request did not meet the audit criteria, the requesting physician was contacted by a senior technologist or attending physician to discuss the necessity for transfusion before proceeding.

As previously mentioned, to evaluate transfusion practice in medical patients during a 5‐year period, analyses included the average number of units transfused per transfusion or clinical specialty, assessment of average Hb levels and evaluation of the proportion of inappropriate transfusions according to prespecified audit criteria (thresholds used for RBC transfusions in clinical practice). A transfusion episode for medical patients was defined as each day a patient was transfused, and the number of units per episode was calculated by adding all units transfused on the same date [10]. The transfusion triggers were defined as follows: a Hb level of <7.0 g/dL for non‐bleeding medical patients without comorbidities and <8.0 g/dL for non‐bleeding medical patients with comorbidities. Comorbidities warranting the higher transfusion threshold of 8.0 g/dL included cancer, haematological malignancies, pneumopathy and cardiovascular disease. Furthermore, higher Hb thresholds were applied in specific clinical scenarios, such as thalassaemia, acute sepsis, active bleeding and hemodynamic instability.

In emergency cases, such as active bleeding or haemodynamic instability, an electronic transfusion request could be processed without the patient's Hb value being a prerequisite for RBC unit release. Episodes of transfusion involving active bleeding or thalassaemia were excluded from the analysis. The term ‘clinic’ refers to either departments or units.

ChatGPT‐5.0 (OpenAI) was employed for language editing.

Statistics

Statistical analyses included the generation of descriptive statistics. Data for RBC units delivered and Hb levels were presented as means ± standard deviations (SDs). Proportions of transfusions with Hb level above and below the thresholds of 7 and 8 g/dL were calculated. Mean Hb levels and proportions above the thresholds of 7 and 8 g/dL were compared over time and by cancer status, presence/absence of other or any comorbidities and hospital department/unit using both parametric and nonparametric tests (chi‐square tests, t‐tests, Wilcoxon tests, one‐way analysis of variance [ANOVA], Kruskal–Wallis tests). Bonferroni tests were used for post hoc analysis following ANOVA. Linear regression, with calendar year treated as a continuous independent variable, was used to assess monotonic (linear) trends in Hb values over time. The Cochran–Armitage trend test was applied to assess monotonic (linear) trends across ordered categories, with the binary response variable indicating whether Hb levels exceeded the thresholds of 7 and 8 g/dL. Despite violations of normality, parametric tests were preferred because of the large sample size. Fixed‐ and random‐effects (random intercept) models, with robust standard errors, were used to assess the association between year and mean Hb levels while adjusting for the other predictors (comorbidity status and hospital department/unit). Fixed‐ and random‐effects logistic regression models were used to assess the association between year and Hb levels above the thresholds of 7 and 8 g/dL while adjusting for relevant predictors. The Hausman test was used to assess whether a random‐effects model (random intercept for hospital departments/units) was acceptable. The squared terms of year and interaction effects were also included in the models as appropriate. When squared terms of year were used, the turning (minimum) point was calculated as follows: coefficient of year/(2 × the coefficient of squared values of year). Calendar year was treated either as a categorical or a continuous variable in the regression models. In the latter case, year values were centred at zero by subtracting 2020, the first year of observations. The Akaike information criterion (AIC) and the Bayesian information criterion (BIC) were calculated for all models.

All statistical analyses were conducted using Stata version 18.0 (StataCorp, College Station, TX, USA). Random‐effects models were applied using the xtreg, re, xtlogit, re and mixed/melogit commands.

RESULTS

Between 2020 and 2024, a total of 23,322 transfusions were recorded in medical patients (range: 4078–5394/year). This count excludes 2037 episodes involving active bleeding or thalassaemia. Half of the transfusions were administered to patients with cancer (n = 11,574 out of 23,322; 49.63%). Approximately one‐fourth of transfusions (n = 5312 out of 23,322; 22.78%) involved patients with comorbidities other than cancer, with annual proportions ranging from 18.81% in 2024 to 26.80% in 2023 (chi‐square p < 0.001). In total, 6436 out of 23,322 transfusion episodes (27.60%) involved patients without comorbidities, ranging from a low 22.49% in 2020 to a high 34.77% in 2023 (chi‐square p < 0.001). The exact number of units administered was available for 22,949 out of 23,322 transfusions (98.4%). The overall mean number of RBC units per transfusion episode was 1.18, with no significant variation across the years (p > 0.05; Table 1).

TABLE 1.

Total episodes of transfusion and mean red blood cell units transfused per transfusion episode, excluding transfusions for active bleeding or thalassaemia (Attikon University Hospital, Athens, Greece; 2020–2024).

	Year—overall	Year 2020	Year 2021	Year 2022	Year 2023	Year 2024	p‐value
Total transfusion episodes	23,322	4144	4559	5394	4078	5147
Mean (±standard deviation) RBC units per transfusion episode	1.18 (±1.51)	1.19 (±0.47)	1.21 (±1.13)	1.20 (±2.89)	1.14 (±0.45)	1.15 (±0.41)	0.11

Note: The p value was derived from one‐way ANOVA.

Abbreviations: ANOVA, analysis of variance; RBC, red blood cell.

Hb levels were available for 23,321 transfusion episodes. The mean Hb level at the time of transfusion over the entire period was 7.18 ± 0.78 g/dL, which was significantly lower (ANOVA and Bonferroni tests p < 0.05) in all years following 2020 (7.28 ± 0.84 g/dL). However, this decreasing trend reversed in 2024, with a slight increase to 7.20 ± 0.76 g/dL (Table 2). The mean Hb level at transfusion time over the entire period was 7.25 ± 0.66 g/dL, 7.24 ± 0.87 g/dL, and 7.00 ± 0.87 g/dL among cancer patients, patients with comorbidities other than cancer and patients without comorbidities, respectively (Table 2).

TABLE 2.

Haemoglobin at transfusion for all episodes and stratified by comorbidity and cancer status, overall and by year (Attikon University Hospital, Athens, Greece; 2020–2024).

	Overall	Year 2020	Year 2021	Year 2022	Year 2023	Year 2024	p value
Mean (standard deviation)
All (n = 23,321)	7.18 (±0.78)	7.28 (±0.84)	7.15 (±0.79)	7.16 (±0.77)	7.12 (±0.75)	7.20 (±0.76)	<0.01
Cancer (n = 11,574)	7.25 (±0.66)	7.30 (±0.69)	7.23 (±0.63)	7.31 (±0.59)	7.12 (±0.70)	7.25 (±0.69)	<0.01
Comorbidities other than cancer (n = 5311)	7.24 (±0.87)	7.35 (±0.99)	7.21 (±0.86)	7.15 (±0.88)	7.25 (±0.80)	7.27 (±0.80)	<0.01
Cancer or other comorbidities (n = 16,885)	7.25 (±0.73)	7.32 (±0.78)	7.22 (±0.71)	7.26 (±0.71)	7.17 (±0.75)	7.26 (±0.72)	<0.01
Without cancer or any other comorbidity (n = 6436)	7.00 (±0.87)	7.14 (±1.01)	6.91 (±0.95)	6.95 (±0.87)	7.01 (±0.75)	7.04 (±0.82)	<0.01
Hb > 7 (n, %)
All	14,733 (63.17%)	2865 (69.14%)	2810 (61.64%)	3303 (61.23%)	2399 (58.83%)	3356 (65.20%)	<0.01
Without cancer or any other comorbidity	3433 (53.34%)	563 (60.41%)	523 (47.72%)	803 (49.23%)	750 (52.89%)	794 (58.43%)	0.25
Hb > 8 (n, %)
All	1426 (6.11%)	345 (8.33%)	527 (5.64%)	302 (5.60%)	212 (5.20%)	310 (6.02%)	<0.01
Cancer or other comorbidities	1118 (6.62%)	269 (8.37%)	198 (5.72%)	227 (6.03%)	150 (5.64%)	274 (7.23%)	0.18
Without cancer or any other comorbidity	308 (4.79%)	76 (8.15%)	59 (5.38%)	75 (4.60%)	62 (4.37%)	36 (2.65%)	<0.01

Note: Transfusion Hb thresholds were ≤8 g/dL for patients with cancer/comorbidities and ≤7 g/dL for patients without comorbidities. Transfusions for active bleeding or thalassaemia were excluded. p‐values were derived from one‐way ANOVA for Hb levels and from the Cochran–Armitage test for trend for the proportions of Hb values above or below the thresholds of 7 and 8 g/dL.

Abbreviations: ANOVA, analysis of variance; Hb, haemoglobin.

The proportion of transfusions with Hb levels above 7.00 g/dL showed a statistically significant decrease over time (Cochran–Armitage test for monotonic trend, p < 0.01), declining from 69.14% in 2020 to 58.83% in 2023 (Table 2). Among patients without any comorbidities, for whom 7.00 g/dL represents the transfusion threshold, nearly half (n = 3433 out of 6436; 53.34%) of the transfusions occurred above this cut‐off. A substantial drop was observed: from 60.41% in 2020 to 47.72% in 2021. However, this decline was not sustained, rising in subsequent years to reach 58.43% in 2024 (Cochran–Armitage test for monotonic trend, p = 0.25) (Table 2). The proportion of transfusions with Hb levels above 8.00 g/dL decreased significantly over the years, from 8.33% in 2020 to 5.20% in 2023 (Cochran–Armitage test for monotonic trend, p < 0.01) (Table 2). Among patients with comorbidities, with 8.00 g/dL as threshold, 6.62% of transfusions exceeded this level.

The mean Hb levels over time by clinic are presented in Figure 1. The lowest overall values were observed in the Emergency Department (6.19 ± 1.43 g/dL; 45 transfusions) and the fourth Department of Internal Medicine (6.99 ± 0.90 g/dL; 4300 transfusions), while the highest values were seen in the Dermatology Department (8.15 ± 1.20 g/dL; 2 transfusion episodes) and the Cardiac Care Unit (7.72 ± 0.86 g/dL, 133 transfusion episodes). Most transfusions (n = 14,831; 63.59%) occurred in patients from the second Department of Internal Medicine, with a mean Hb level of 7.22 ± 0.72 g/dL.

FIGURE 1.

Mean haemoglobin (Hb) at transfusion for all episodes of transfusion administered to patients treated at Attikon University Hospital, Athens, Greece, stratified by hospital clinic and year (2020–2024). Transfusions for active bleeding or thalassaemia were excluded.

Table S1 presents mean Hb levels by hospital clinic for patients with and without comorbidities, along with the proportions of transfusions administered above 8 g/dL for patients with comorbidities and above 7 g/dL for those without. Notable differences were observed across hospital departments/units. For example, markedly high values were recorded in the Cardiac Care Unit, where the mean Hb level over the 5‐year period was 7.87 ± 0.91 g/dL for patients with comorbidities and 7.41 ± 0.68 g/dL for those without. In contrast, in the Critical Care Unit, a clinic of similar nature, the corresponding mean Hb levels were lower: 7.15 ± 0.88 g/dL and 7.06 ± 0.81 g/dL for patients with and without comorbidities, respectively.

Table S2 presents the results of the linear regression models (Hb levels treated as a continuous dependent variable), which support a decreasing trend over time, adjusted for comorbidity status and hospital clinic, along with significant differences among the various hospital clinics. However, the squared term for year had a statistically significant positive coefficient (fixed‐effects: 0.02; 95% confidence interval [CI]: 0.01–0.03), indicating that the decreasing trend flattened or even reversed. Figure 2 shows the results of the fixed‐effects logistic regression model using Hb levels above or below the threshold of 7 g/dL (dichotomous variable) for patients without comorbidities. The odds ratio for year (as a continuous variable) was 0.70 (95% CI: 0.61–0.80), while the corresponding exponentiated coefficient for the squared term of year was 1.10 (95% CI: 1.06–1.13). In terms of interpretation, over the calendar years and controlling for hospital department/unit, the odds of having Hb levels >7 g/dL at transfusion among patients without comorbidities initially decreased by about 30% per year. However, this decreasing trend gradually flattened, indicated by a 10% increase in odds per unit increase in the square of the year. Using the logistic regression coefficients (the natural logarithms [ln] of the odds ratios), the estimated turning point (minimum) was around 2022, calculated as −ln (0.70)/(2 × ln [1.10]). Similar results from logistic regression models, applied to transfusions with Hb levels above or below 8 g/dL (dichotomous variable) among patients with comorbidities, are shown in Figure 3.

FIGURE 2.

Odds ratios and 95% confidence intervals (CIs) from fixed‐effects logistic regression among patients without comorbidities treated at Attikon University Hospital, Athens, Greece, during the period 2020–2024 (n = 6434). The proportion of transfusions with haemoglobin (Hb) levels above 7 g/dL (log‐odds) was modelled against year (treated as a continuous variable centred at zero, i.e., minus 2020), the squared term of year and hospital department/unit. The odds ratio for year represents the change per additional year. Odds ratios for departments/units were calculated relative to the second Department of Internal Medicine, which served as the reference group. Transfusions given for active bleeding or thalassaemia were excluded.

FIGURE 3.

Odds ratios from fixed‐effects logistic regression among patients with comorbidities treated at Attikon University Hospital, Athens, Greece, during the period 2020–2024 (n = 16,886). The proportion of transfusions with haemoglobin (Hb) levels above 8 g/dL (log‐odds) was modelled against year (treated as a continuous variable centred at zero, i.e., year minus 2020), the squared term of year and hospital department/unit. The model includes interaction effects between year and department/units. The odds ratio for year represents the change per additional year. Odds ratios for departments/units were calculated relative to the second Department of Internal Medicine, which served as the reference group. Transfusions given for active bleeding or thalassaemia were excluded.

DISCUSSION

Five years after the initial evaluation of implementing the electronic transfusion request system for the first time in a Greek tertiary hospital [8], a moderate compliance with the restrictive transfusion strategy was observed. Regarding Hb thresholds used for RBC transfusions, the results indicate a decreasing trend over time, after adjusting for comorbidity status and hospital clinic, with significant differences among the various clinics. This decreasing trend in transfusions above prespecified Hb thresholds across most clinics later plateaued and eventually reversed, yet the proportion remained at a lower level than the initial point.

Blood transfusion is one of the most overused therapeutic procedures, applied inappropriately and exhibiting significant variation both within and between organizations. Despite the numerous guidelines published, this inconsistency in transfusion strategies has persisted, contrary to expectations of transfusion practice standardization [11]. Based on these findings, PBM programmes have developed and promoted the use of lower Hb levels by having the patient as the central focus rather than the Hb level [12]. In most cases, it is impossible for all approaches supporting PBM to be equally or efficiently applied within a single organization, making PBM performance comparisons among several institutions or countries difficult [13].

PBM programmes have been increasingly integrated into routine clinical practice across many European countries. However, the implementation level can differ significantly between countries and regions, with some having national policies promoting PBM while others are in early stages of adoption [14]. For instance, in Greece there is no systematic effort to disseminate and implement PBM in hospitals. Furthermore, a major challenge for the implementation of PBM in Greece is the absence of national transfusion guidelines. Under these circumstances, the TMC of Attikon University Hospital decided, after the establishment of local transfusion guidelines, to apply the most practical and potentially efficient aspect of PBM, that is, the implementation of an electronic transfusion prescribing tool, including the patient's diagnosis, the most recent laboratory values, transfusion indication and other medical conditions that affect transfusion decisions to ensure the best clinical outcome. Based on a retrospective audit, our data suggest that this approach has been steadily increasing in acceptance and implementation over time among the hospital's medical staff. However, a reduction in blood utilization has not been achieved in our case. While stronger adherence to the restrictive transfusion strategy was expected—particularly when compared with results reported in the current literature in similar clinical settings and in the absence of PBM programme implementation [2, 3, 4]—this was not observed. Several reasons might account for this, such as the varying number of patient admissions over time and the hospitalization of many COVID‐19 patients, which limited the operation of other clinics during the COVID‐19 pandemic. Similar to most reforms, the initial adoption of PBM occurred somewhat gradually and irregularly [8].

Despite the steady improvement in the results of the programme for a relatively long period of about 3 consecutive years, a reversing trend was noted during the last years of the follow‐up period. Although potential random variability could be implicated, it does not appear to be a random result because this finding was supported by the analysis of most statistical parameters. Several factors may have contributed to the recently observed decrease in adherence to restrictive transfusion practices. For instance, the end of the pandemic's effects, which has led to an increasing number of potential blood donors and consequently to a relatively increased blood supply, might have reduced the need for strict control of transfusion demands. However, in our opinion, the most critical factor contributing to the weakening of the restrictive transfusion policy is perhaps the acute reduction in blood bank staff that occurred during the last year of the follow‐up period. This reduction caused significant difficulties in manually reviewing and discussing with clinicians the appropriateness of all transfusion requests that were not aligned with the restrictive strategy. Since contact with the clinicians for all ‘inappropriate’ transfusion orders entailed considerable workload by the insufficient transfusion service staff, the consistent commitment to this intervention was not always feasible. Moreover, the lack of ongoing education on the benefits of a restrictive transfusion policy, combined with clinicians' reluctance to comply with the restrictive approach or accept interventions that modify their clinical judgement, has probably also resulted in the negative findings of the past year.

Considering all stated issues, a potential substantial improvement could be the addition of a mandatory clinical decision‐support tool to the electronic blood ordering system. Such a tool, besides alerting clinicians to inappropriate orders, could facilitate auditing by the blood bank staff with less additional workload. Clinical decision tools have been previously shown to reduce the occurrence of inappropriate blood transfusions [15, 16, 17, 18]. This has been achieved in both paediatric and adult hospitals [19, 20].

As far as Hb transfusion thresholds are concerned, although a similar pattern of adherence to restrictive transfusion cut‐offs was observed across most clinics over time, significant differences were noted among them. The overall mean transfusion Hb values steadily decreased over the period, but this trend reversed at the end of the follow‐up in most clinics. Conversely, the Critical Care Unit demonstrated more sustainable compliance with the restrictive practice. Of note, the highest values recorded in the Cardiac Care Unit were probably due to the reluctance of cardiologists to adopt the lower Hb values of the restrictive strategy. This reluctance appears to be partially justified, considering recent evidence suggesting that ‘liberal’ rather than ‘restrictive’ transfusion strategies are associated with significantly improved safety for hospitalized patients with stable cardiovascular disease and/or acute coronary syndromes [21]. Therefore, a more liberal approach to managing anaemia may be warranted in high‐risk patients [22], such as those with acute coronary syndromes. However, the electronic blood ordering system used in our hospital, by incorporating elements like clinical status and comorbidities affecting oxygen‐carrying capacity—such as coronary artery disease, unstable angina, hemodynamic instability, respiratory disease, presence of malignancy or bleeding—allows a ‘patient‐based’ transfusion practice rather than a single numerical transfusion trigger approach.

Several limitations of our study have to be acknowledged. As a single‐centre study, its generalizability is limited. In statistical terms, there were violations of the normality assumption regarding Hb levels. However, parametric tests were applied, as their robustness to non‐normality is generally acceptable with large sample sizes. Additionally, because of the lack of a systematic effort to adopt a restrictive transfusion policy prior to PBM implementation and the limited data available for assessing blood utilization during the pre‐PBM period, establishing a definitive causal relationship between the local PBM and the adoption of a restrictive transfusion trigger is not feasible. However, considering that during the pre‐PBM era there was no systematic approach or effort to review the appropriateness of transfusion requests, it is evident that implementing a prospective audit of transfusion demands through an electronic transfusion request system—taking into account individualized laboratory and clinical patient data—along with the adoption of a restrictive RBC transfusion strategy, represents a positive step towards the optimal use of blood resources. On the other hand, the main strength of our study is the extended study period, which provides a comprehensive understanding of the effects of the local PBM system on blood utilization and allows for the identification of several system weaknesses, thereby facilitating the implementation of corrective measures.

A thorough overview of national PBM policies across Europe shows that the guidelines are limited, varied and often inconsistent [23, 24, 25]. Despite the World Health Organization's (WHO) endorsement of PBM since 2010, there remains a significant gap in its widespread adoption. Some countries have advanced in areas such as anaemia treatment and reducing transfusions, but most healthcare systems lack clear, cohesive national policies to fully implement the three core principles of PBM, that is, optimizing RBC levels, minimizing blood loss and improving patient tolerance to anaemia. In response, the WHO has launched a new initiative to develop comprehensive PBM implementation guidelines, aiming to overcome these challenges and foster a patient‐centred approach to blood management throughout Europe [26].

Conclusively, effective collaboration with clinicians is a critical element in the optimal utilization of blood products. Successful cooperation requires consistent and long‐term education, steady feedback and positive interaction. Unfortunately, this is not always the case. Several challenges, such as staff shortages, increased workload, lack of national strategies to support PBM‐related activities and physicians' adherence to ingrained institutional transfusion habits, may impede the full implementation of established transfusion guidelines. Thanks to the efforts of the blood bank staff and the effective collaboration with the majority of physicians, a gradual improvement in adherence to the restrictive transfusion strategy has been observed over time at Attikon University Hospital. Although other aspects of a comprehensive PBM programme—such as the optimization of the patient's own blood production, the implementation of preoperative anaemia management or the enhancement of haemostasis—have not been applied in our setting, the observed improvement in RBC transfusion practices was nonetheless significant. Once a clinical decision‐support tool is fully integrated with the electronic blood ordering system, then PBM could be extended to incorporate algorithms for the investigation and management of anaemia, the optimization of haemostasis and the implementation of measures to reduce blood utilization during surgery. Effective monitoring, analysis and feedback demand considerable time and personnel resources. Therefore, a key prerequisite for reaching these objectives is ensuring sufficient staffing in blood bank units and the development of a comprehensive national legal framework that endorses PBM programmes.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Table S1. Mean haemoglobin (Hb) levels at time of transfusion for all episodes, analysed for patients with comorbidities (CO; cancer or any other comorbidity) and without comorbidities (NC), stratified by hospital clinic year (2020–2024). Proportions of transfusions above 8 g/dL in patients with comorbidities and above 7 g/dL in patients without comorbidities are also presented. Transfusions for active bleeding or thalassaemia were excluded from the analysis.

Table S2. Regression models (coefficients and 95% confidence intervals [CIs]) for haemoglobin levels against year, presence of comorbidities and hospital department/unit. Transfusions for active bleeding or thalassaemia were excluded. *p < 0.05.

Nikolopoulos GK, Koutoulogeni A, Kyriakou E, Tsalas S, Papandreadi A, Douramani P, et al. Evaluating compliance with restrictive blood transfusion practices: A 5‐year study from a Greek tertiary hospital. Vox Sang. 2026;121:322–330.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.