Investigating the association between blood transfusion and clinical outcomes in patients with acute coronary syndromes: a data linkage approach to Patient Blood Management

Abstract

Background

The appropriateness of the use of blood transfusion in patients with acute coronary syndromes (ACS) remains contested. In general, studies addressing this issue were based on data from clinical trials, registries, or electronic medical records, and were conducted across different settings. Our study aimed to use a linked patient blood management data system from existing hospital databases to examine the association between blood transfusion and in-hospital mortality, length of stay (LOS) and readmission rates among patients with ACS, and to investigate this relationship at different haemoglobin (Hb) concentrations.

Materials and methods

This was a retrospective observational study of patients admitted to participating hospitals between 1^st January 2014 to 31^st December 2017 with ACS recorded as primary diagnosis. Admission and nadir Hb concentrations were categorised as ≤100 g/L and >100 g/L. Generalised estimating equations were used to investigate the association between transfusion and clinical outcomes, while accounting for the correlation of multiple admissions from the same patients across hospitals over the study period.

Results

Of the 9,952 admissions included, blood transfusions occurred in 705 (7.1%). In unadjusted analyses, transfusion was associated with an increased risk of in-hospital mortality (OR: 2.97; 95% CI: 2.14–4.13) and an average LOS 3.55 (95% CI: 3.38–3.72) times longer. After adjusting for demographic and clinical factors, transfusion was associated with an increased risk of in-hospital mortality when Hb >100 g/L. Transfusion was not associated with the risk of readmission.

Discussion

The effect of transfusion on in-hospital mortality was largely dependent on the pre-transfusion Hb concentration. When Hb was >100 g/L transfusion was associated with increased mortality, whereas when Hb ≤100 g/L no association was observed.

INTRODUCTION

Anaemia is an independent risk factor for mortality and major adverse cardiovascular outcomes in patients with acute coronary syndromes (ACS)^1,2. In this setting, blood transfusion can be beneficial by increasing oxygen delivery to the vulnerable myocardium and reducing ischemic symptoms³. However, inappropriate transfusion practices may lead to circulatory overload and increase thrombogenicity, hence, worsening clinical outcomes^4,5. Furthermore, data suggest that red blood cells (RBC) are rapidly depleted of nitric oxide during storage, which may affect oxygen delivery^6,7. The appropriateness of the use of blood transfusion in the ACS population remains contested. Recent meta-analyses have shown an association between transfusion and increased mortality^4,8,9. However, several studies have focused on the adequate transfusion threshold at which to maximise benefit and minimise harm^2,10–14. While some found an association between RBC transfusion and increased mortality in patients with a nadir haemoglobin (Hb) concentration >100 g/L^10,12, others reported decreased mortality with an admission Hb concentration ≤110 g/L¹⁴ or < 120 g /L². In general, these studies were based on post hoc analysis of data from clinical trials^2,12, registries^10,14 or electronic medical records (EMR)¹³ focusing on mortality and were conducted in heterogeneous cohorts.

Patient blood management (PBM) is an evidence-based patient-oriented approach to limit the use and need for blood transfusions and improve clinical outcomes^15,16. PBM has been adopted by a number of departments, hospitals and health networks worldwide and has been shown to provide robust information to monitor blood use and transfusion practices^15,17,18. However, little information exists on PBM implementation in the setting of ACS. The aim of this study was to use a linked PBM data system from existing administrative hospital databases to examine the association between blood transfusion and in-hospital mortality, length of stay (LOS) and readmission rates among patients with ACS, and to investigate this relationship at different Hb concentrations.

MATERIALS AND METHODS

Study population

This was a retrospective observational study of adult patients (aged >18) admitted to participating hospitals between 1^st January 2014 to 31^st December 2017 with a record for ACS as primary diagnosis. ACS was defined according to the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM), comprising of codes I20–I24¹⁹. Chronically transfused patients (n=13)²⁰ were excluded from the study, as were admissions without pre-transfusion Hb (n=34) and those with massive transfusions (n=19)²¹, defined as a transfusion of at least 5 or 10 units of RBC in 4 or 24 hours, respectively. Due to the additional risks of complications associated with transfusion in these populations, separate guidelines apply to these populations.

Data source

This study was undertaken across two Local Health Districts (LHDs) in New South Wales (NSW) and utilised routinely-collected longitudinal data from 4 hospitals, all of which operate an EMR which enables clinicians to create electronic orders. Ethics approval for the study was granted by the Human Research Ethics Committee of the relevant Local Health District.

Data were extracted from three hospital clinical information systems including the Patient Administration System (PAS), which contains data on hospital admissions, Laboratory Information System (LIS), which contains data on test utilisation, and the Blood Bank database (BB) which provides data on blood product events. Information recorded in these databases includes demographics, diagnoses, in-hospital procedures, mode of separation, test results date-time, electronic dispensing date-time of each blood product from the BB, and units of blood products transfused.

Data quality assessment and data linkage

A rigorous data quality assessment process was employed to each dataset to evaluate the accuracy, completeness, consistency, relevance, timeliness, uniqueness, and validity of the data sources²². This process included the identification of missing data, duplicates, and formatting issues. The linkage of hospital inpatient, pathology laboratory and blood bank data was undertaken using non-identifiable patient medical record number, gender, and date of birth, common to all datasets. Records were considered “linked” if there was an exact match on all identifiers. Because a given patient could have multiple visits at the same or different sites overtime, only test results and blood transfusions date-times which fell between admission and discharge date-times were considered.

Definitions and outcome measures

• Transfusion date-time. In the BB system, the transfusion event is automatically added by the system 30 minutes after the blood product being dispensed for transfusion. For this reason, in our analysis, the dispense date-time was used as a proxy for transfusion date-time.

• Charlson Comorbidity Index (CCI). This is among the best-known and most widely used indexes of comorbidity with administrative data and it has been shown to be an appropriate prognostic indicator of in-hospital mortality in ACS patients²³. For this analysis, the updated CCI was used²⁴ and categorised as “0” (no comorbidity), “1–3”, “≥4”.

• Bleeding. Bleeding events were defined as per the Thrombolysis in Myocardial Infarction (TIMI) trial²⁵. According to TIMI bleeding classifications, major bleeding is associated with either intracranial haemorrhage or a haemoglobin decrease of 50 g/L or more. Minor bleeding is associated with a haemoglobin decrease between 30 g/L and 50 g/L. For this analysis, available consecutive Hb test results were used to derive such events and the occurrence of bleeding was considered as a binary variable, regardless of its severity.

• Hb measurements. Admission Hb was defined as the first Hb test result available after admission, whereas nadir Hb was defined as the lowest Hb test result occurring during hospital stay. In the presence of bleeding or transfusion events, only the first occurrence was considered, and the nadir level before the event was measured. For the purpose of this analysis, Hb values were considered as categorical variables (“≤100 g/L”, “>100 g/L”) to be in line with current transfusion practices recommendations among the ACS population²⁰.

• In-hospital procedures. The Australian Classification of Health Interventions (ACHI), Tenth Edition²⁶, was used to identify percutaneous coronary intervention (PCI) and coronary artery bypass graft (CABG) procedures performed during hospitalisation. Description of the codes used are provided in a supplementary table.

• Outcome measures. The primary outcome was in-hospital mortality. Secondary outcomes were 28-day readmission and hospital length of stay (LOS). The former was defined as readmission to any participating hospital within 28 days of the discharge date, excluding transfers, while the latter was calculated as the length of time between hospital admission and discharge.

Statistical analysis

Hospital admissions were categorised according to the occurrence of transfusion. Baseline characteristics and univariate associations with clinical outcomes were compared using Chi-square tests for categorical variables and nonparametric Kruskal-Wallis tests for continuous variables. Unadjusted odds ratios, overall and stratified by admission and nadir Hb categories, were used to measure the strength of association between blood transfusion and categorical outcomes (in-hospital mortality and 28-day readmission). For LOS, generalised linear regression models²⁷, assuming a normal distribution with a log link function, were used to evaluate the unadjusted relationship with blood transfusion. Due to this transformation, the parameter estimates generated in this model represent the multiplicative effect on LOS of transfusion occurrence.

Pearson correlation was used to assess the linear relationship between admission and nadir Hb, both considered as continuous variables. The coefficient of 0.87 (p-value <0.001) indicated strong correlation between them and, consequently, would lead to collinearity if both variables are included. In the presence of collinearity, estimated coefficients are less precise, more sensitive to small changes in the model, and tests of significance of predictors are inaccurate²⁸.

To evaluate the independent relationship between transfusion and clinical outcomes, multivariate regression models were built using generalised estimating equations (GEE)²⁹. The appropriateness of the GEE approach relies on the fact that it takes into account the correlation of multiple admissions from the same patients across hospitals over the study period. Therefore, logistic regression was used for in-hospital mortality and 28-day readmission, whereas normal regression with log link was used for LOS. Models were adjusted for CCI, age, gender, in-hospital procedures (PCI, CABG), bleeding events, hospital, and also accounted for admission and nadir Hb separately. Since the effect of blood transfusion is expected to be dependent on the Hb level^12,14, an interaction term between them was also considered.

Due to the influence of CABG on transfusion practice and mortality, sensitivity analyses were conducted by excluding admissions where CABG occurred, with further limitation to the elderly population (aged ≥65). Statistical analyses were conducted using SAS software version 9.4 (SAS Institute Inc, Cary, NC, USA). A significance level of 5% was used.

RESULTS

This study included 9,952 admissions across 4 hospitals with data on pre-transfusion Hb tests, transfusion, and outcomes. Of these, blood transfusions occurred in 705 (7.1%) admissions. Table I presents the baseline characteristics of admissions according to the occurrence of transfusion. Compared with admissions without transfusion, those in which a blood transfusion occurred more likely involved older patients, more often females, and patients with more comorbidities. These patients also had more in-hospital events, including bleeding and CABG, and less frequent PCIs. The median admission and nadir Hb levels were 137 g/L and 131 g/L, respectively, for admissions without transfusion, and 117 g/L and 101 g/L, respectively, when transfusion occurred. Among the latter, single-unit transfusions accounted for 38%, followed by those using 3 or more RBC units (32%), and 2-unit transfusions (30%).

Table I.

Baseline characteristics of hospital admissions with and without transfusion^*

Characteristics	No transfusion (n=9,247)	Transfusion (n=705)	p-value
Demographics
Age, ≥65 years	5,631 (61)	538 (76)	<.0001
Gender, Female	2,987 (32)	256 (36)	0.03
CCI			<.0001
0	6,014 (65)	248 (35)
1–3	2,831 (31)	343 (49)
≥4	402 (4)	114 (16)
In-hospital events
Bleeding (severe or moderate)	345 (4)	158 (22)	<.0001
PCI	2,594 (28)	77 (11)	<.0001
CABG	412 (4)	402 (57)	<.0001
Hb, median (IQR), g/L
Admission	137 (125–149)	117 (96–136)	<.0001
Nadir	131 (117–143)	101 (84–128)	<.0001
RBC units transfused
1	NA	266 (38)	NA
2		211 (30)
≥3		228 (32)

^*Data presented as n (%), unless specified.

CCI: Charlson comorbidity index; Hb: haemoglobin; PCI: percutaneous coronary intervention; CABG: coronary artery bypass graft; RBC: red blood cell; NA: not applicable.

Furthermore, Table II shows that transfusion was significantly associated with increased LOS (12.3 days vs 3.2 days; p-value <0.001) and higher in-hospital mortality rates (7 vs 2%; p-value <0.001) when compared with admissions without transfusion. There was not an empirical basis to suggest an association between transfusion and 28-day readmissions (12 vs 10%; p-value=0.053).

Table II.

Clinical outcomes among hospital admissions with and without transfusion

Outcomes	All admissions (n=9,952)	No transfusion (n=9,247)	Transfusion (n=705)	p-value
LOS, median (IQR), days	3.4 (1.8–6.7)	3.2 (1.7–5.9)	12.3 (8.2–19.5)	<0.001
28-day readmission, n (%)	978 (11)	894 (10)	84 (12)	0.053
In-hospital mortality, n (%)	258 (3)	212 (2)	46 (7)	<0.001

LOS: length of stay; n: number.

The boxplots in Figure 1 show the variation in admission and nadir Hb levels among admissions with transfusion for a total of 687 patients. Due to the longitudinal nature of the data, a patient may have been admitted and transfused more than once, hence the sum of patients above and below 100 g/L for either admission or nadir Hb may not be the same as the total number of patients. Among admissions with Hb levels ≤100 g/L, the median Hb was 89 g/L and varied from 58 to 100 g/L, whereas that of those greater than 100 g/L was 129 g/L, ranging from 101 to 188 g/L. The median nadir Hb prior to transfusion was 84 g/L among admissions with Hb levels ≤100 g/L, and 128 g/L among those with Hb levels over 100 g/L. The range of both nadir Hb level cohorts were the same as those on admission. Table III shows the unadjusted rates of LOS, 28-day readmissions and in-hospital mortality among admissions with and without transfusion. For LOS and in-hospital mortality, the rates were significantly higher for admissions in which transfusion occurred. Blood transfusion was associated with an average LOS 3.55 (95% confidence interval [CI]: 3.38–3.72) times longer than that of admissions without transfusion, and an odds ratio for in-hospital death of 2.97 (95% CI: 2.14–4.13).

Figure 1.

Variation in admission and nadir Hb levels among all admissions with transfusion

Table III.

Univariate analysis, by admission and nadir Hb categories, on associations between blood transfusion and clinical outcomes

	LOS*	28-day readmission†	In-hospital mortality†
Overall	3.55 (3.38–3.72)	1.26 (0.996–1.60)	2.97 (2.14–4.13)
Admission Hb
≤100	1.98 (1.62–2.41)	0.95 (0.60–1.52)	0.99 (0.52–1.87)
>100	3.97 (3.78–4.17)	1.07 (0.79–1.45)	2.95 (1.98–4.42)
Nadir Hb
≤100	1.77 (1.55–2.01)	0.85 (0.60–1.21)	1.12 (0.72–1.73)
>100	3.90 (3.71–4.11)	1.07 (0.74–1.53)	2.35 (1.34–4.11)

^*Multiplicative effect on LOS of transfusion occurrence (95% CI).

^†Odds ratio (95% CI) of transfusion occurrence.

CI: confidence interval; Hb: haemoglobin; LOS: length of stay.

The unadjusted analysis stratified by admission and nadir Hb categories also showed an increase LOS in favour of blood transfusion, and even longer LOS when Hb levels were over 100 g/L. For in-hospital mortality, the transfusion effect when either admission or nadir Hb were 100 g/L or less was not significant. However, when admission or nadir Hb levels were higher than 100 g/L, transfusion was associated with a significant odds ratio for in-hospital death of 2.95 (95% CI: 1.98–4.42) and 2.35 (95% CI: 1.34–4.11), respectively. The unadjusted effects of blood transfusion on 28-day readmission rates were not statistically significant.

The adjusted association between hospital admissions with and without blood transfusion and clinical outcomes, by admission and nadir Hb categories is shown in Table IV. For LOS, although the interaction between transfusion by Hb categories at either admission or nadir was not significant, the occurrence of transfusion was significantly associated with increased LOS. For in-hospital mortality, the interaction between transfusion by Hb categories was significant such that there were no effects of transfusion on the odds ratio of in-hospital death when either admission or nadir Hb levels were 100 g/L or less. However, at admission or nadir Hb higher than 100 g/L, transfusion was significantly associated with an odds ratio of 3.72 (95% CI: 2.26–6.14) and 5.66 (95% CI: 2.59–12.36) for in-hospital mortality, respectively. The adjusted effects of blood transfusion on 28-day readmission rates were not statistically significant.

Table IV.

Adjusted model results, by admission and nadir Hb categories, on associations between blood transfusion and clinical outcomes

Variables	LOS		28-day readmission		Mortality
	Admission Hb	Nadir Hb	Admission Hb	Nadir Hb	Admission Hb	Nadir Hb
CCI
0	Reference		Reference		Reference
1–3	1.45 (1.32–1.58)	1.37 (1.27–1.47)	1.68 (1.39–2.03)	1.64 (1.36–1.98)	3.69 (2.63–5.16)	3.34 (2.38–4.69)
≥4	2.34 (1.99–2.75)	2.04 (1.76–2.36)	2.45 (1.76–3.41)	2.32 (1.65–3.25)	11.04 (7.36–16.57)	9.26 (6.10–14.07)
Hospital
Hospital 1	Reference		Reference		Reference
Hospital 2	0.97 (0.85–1.11)	0.94 (0.84–1.06)	1.19 (0.99–1.43)	1.20 (1.00–1.44)	1.51 (1.04–2.17)	1.56 (1.08–2.24)
Hospital 3	0.70 (0.61–0.80)	0.72 (0.63–0.82)	0.87 (0.68–1.12)	0.88 (0.68–1.13)	0.96 (0.60–1.54)	1.01 (0.63–1.61)
Hospital 4	1.26 (1.14–1.40)	1.23 (1.11–1.37)	0.67 (0.54–0.83)	0.66 (0.53–0.83)	1.54 (1.06–2.24)	1.57 (1.08–2.28)
Age
≥65	Reference		Reference		Reference
<65	0.95 (0.84–1.06)	0.95 (0.86–1.06)	0.87 (0.74–1.02)	0.88 (0.75–1.03)	0.29 (0.18–0.45)	0.30 (0.19–0.48)
Gender
Male	Reference		Reference		Reference
Female	0.97 (0.86–1.09)	0.98 (0.88–1.09)	1.10 (0.94–1.28)	1.09 (0.93–1.27)	1.33 (1.02–1.72)	1.30 (1.00–1.69)
In-hospital events
Bleeding (severe or moderate)	1.03 (0.89–1.20)	1.28 (1.09–1.52)	1.03 (0.65–1.64)	1.07 (0.67–1.70)	1.05 (0.45–2.44)	1.03 (0.42–2.54)
PCI	1.08 (0.99–1.18)	1.12 (1.02–1.23)	1.25 (1.01–1.56)	1.25 (1.00–1.55)	0.82 (0.58–1.16)	0.82 (0.58–1.16)
CABG	1.85 (1.55–2.21)	1.92 (1.64–2.25)	0.70 (0.46–1.08)	0.63 (0.41–0.97)	0.20 (0.09–0.45)	0.17 (0.07–0.42)
Transfusion * Hb
≤100	1.58 (1.29–1.92)	1.48 (1.27–1.71)	0.91 (0.47–1.78)	0.81 (0.51–1.29)	0.86 (0.43–1.69)	1.07 (0.66–1.73)
>100	2.06 (1.90–2.22)	1.75 (1.62–1.90)	1.10 (0.76–1.58)	1.34 (0.86–2.09)	3.72 (2.26–6.14)	5.66 (2.59–12.36)

^*Multiplicative effect on LOS of transfusion occurrence (95% CI).

^†Odds ratio (95% CI) of transfusion occurrence.

CI: confidence interval; LOS: length of stay; PCI: percutaneous coronary intervention; CABG: coronary artery bypass graft.

Sensitivity analyses performed on admissions in which CABG did not occur, and with further limitation to older adults produced similar results. When admission or nadir Hb levels were higher than 100 g/L, transfusion was associated with a significant odds ratio for in-hospital mortality of 3.80 (95% CI: 2.25–6.42) and 6.10 (95% CI: 2.48–14.98), respectively, after excluding CABG, and 2.91 (95% CI: 1.63–5.18) and 4.00 (95% CI: 1.38–11.61), respectively, among those aged 65 and over without CABG.

DISCUSSION

Routinely-collected longitudinal data was used in this large, multi-centre linked PBM data system to investigate the relationship between blood transfusion and clinical outcomes in patients with ACS. The results of our study showed that blood transfusion was associated with increased in-hospital mortality and that the degree of this association varied by Hb concentration. When either admission or nadir Hb was above 100 g/L, we found an association between transfusion and increased risk of in-hospital mortality, whereas the findings did not suggest an association at admission or nadir Hb below 100 g/L.

This finding is broadly consistent with the results of other studies in this area linking blood transfusion with in-hospital mortality without stratification by Hb thresholds^30,31. In a large retrospective cohort of patients admitted with non-ST-segment elevation ACS, Yang et al.³⁰ found that blood transfusion was associated with an increased risk of in-hospital mortality of 1.67 (95% CI: 1.48–1.88). Another multicentre observational study of anaemic patients undergoing PCI for myocardial infarction also found that blood transfusion was associated with an increased risk of in-hospital mortality of 2.02 (95% CI: 1.47–2.79)³¹. Similarly, Sherwood et al.³² reported an increased risk of in-hospital mortality in patients undergoing PCI who received post-procedure transfusion (odds ratio [OR]: 4.63; 95% CI: 4.57–4.69). However, a small number of studies exist that indicate the contrary^33,34. In a French cohort of patients admitted with acute myocardial infarction, transfusion was not associated with in-hospital mortality³⁴. Also, a large multi-centre study using EMR data from hospitalised patients suggested an association between transfusion and decreased in-hospital mortality¹³ (OR: 0.73; 95% CI: 0.58–0.92).

Studies on the efficacy of blood transfusion on mortality according to different Hb concentrations have generated conflicting evidence. In a large retrospective cohort study using data from the same database used by Yang et al.³⁰, Alexander et al.¹⁰ investigated the relationship between transfusion and in-hospital mortality by different nadir Hb thresholds. Among patients with nadir Hb >100 g/L, transfusion was associated with an increased risk of in-hospital mortality (OR: 3.47; 95% CI: 2.30–5.23), whereas a statistically significant association between the same among patients with a nadir at lower thresholds was not found. Similarly, using data from three clinical trials, Rao et al.¹² reported an association between transfusion and an increased 30-day mortality at nadir Hb >100 g/L, but no association at nadir Hb ≤80 g/L. Additionally, a recent meta-analysis of observational studies in ACS patients indicated an association of transfusion with an increased relative risk of mortality at Hb >100 g/L of 3.34 (95% CI: 2.25–4.97)⁸. A large prospective observational study of older patients admitted for ACS also supported these findings¹¹. Conversely, Wu et al.¹⁴ found that transfusion was associated with a reduction in 30-day mortality among older patients with admission Hb ≤110 g/L.

To our knowledge, there is little evidence about the association between transfusion and LOS or readmission rates. The available information is either from univariate associations^13,30 or heterogeneous clinical populations^35–38. Nonetheless, our findings also suggest an association between transfusion and increased LOS^{13,30,36–38} and no association with the risk of readmission^35,37.

In-hospital mortality rates, LOS and readmissions have been shown to be important indicators of the efficiency of hospital management, patient quality of care and functional evaluation^39–41. Our linked PBM data system using routinely-collected hospital data has served as a valuable tool to produce evidence which support findings of previous meta-analyses and observational studies on the effects of transfusion on clinical outcomes. More importantly, this data linkage approach has demonstrated that, regardless of the index pre-transfusion Hb concentration (admission or nadir) considered, transfusion at Hb >100 g/L was associated with increased mortality among ACS patients, which is in line with current transfusion practice recommendations by the National Blood Authority²⁰. Finally, this tool can be useful for evidence validation, ongoing monitoring of transfusion practice, clinical outcomes and for national or international benchmarking.

Study limitations

There are some limitations to our study. Firstly, from a methodological perspective, observational study designs produce findings that are vulnerable to different types of bias and other residual (unmeasured) factors which can potentially confound study outcomes. As such, a cause-and-effect relationship between transfusion and clinical outcomes cannot be established. Secondly, due to the administrative nature of the data and its built-in limitations, appropriate cohort selection is dependent on the accuracy of the registration and the validity of the coding of diagnoses and in-hospital procedures. As such, the location where the nadir Hb occurred and its temporal correlation with transfusion are not available, and Type 2 myocardial infarction episodes may not have been excluded. Additionally, data on medical history, medication and cardiac biomarkers were not available. Finally, our data did not capture patient activity outside of participating hospitals, as such, readmission rates may have been underestimated.

CONCLUSIONS

Blood transfusion in patients with ACS was associated with an increased risk of in-hospital mortality and longer LOS but it was not associated with readmission rates. The effect of transfusion on mortality was largely dependent on the pre-transfusion Hb concentration, such that when Hb was >100 g/L transfusion was associated with increased in-hospital mortality, whereas when Hb ≤100 g/L no association was observed. Until randomised controlled trials of appropriate transfusion strategies are conducted, our linked PBM approach using administrative data suggest caution against the use of blood transfusion in patients with higher Hb concentrations.