Effect of four hemoglobin transfusion threshold strategies in patients with acute myocardial infarction and anemia: a target trial emulation using MINT trial data

Abstract

Background

The specific hemoglobin threshold to guide red blood cell (RBC) transfusion for patients with acute myocardial infarction (MI) and anemia is uncertain.

Objective

To estimate the efficacy of four individual hemoglobin thresholds (<10g/dL [<100g/L], <9g/dL [<90g/L], <8g/dL [<80g/L], and <7g/dL [<70g/L]) to guide transfusion in patients with acute MI and anemia.

Design

Pre-specified secondary analysis of the Myocardial Ischemia and Transfusion (MINT; ClinicalTrials.gov number: NCT02981407) trial using target trial emulation methodology.

Setting

144 clinical sites in 6 countries.

Participants

3,492 MINT trial participants with acute MI and hemoglobin level <10g/dL.

Interventions

Four transfusion strategies to maintain patients’ hemoglobin concentration at or above thresholds of 10, 9, 8, or 7g/dL. Protocol exceptions were permitted for specified adverse clinical events.

Measurements

Data from the MINT trial was leveraged to emulate four transfusion strategies and estimated per-protocol effects on the composite outcomes of 30-day death or recurrent MI (death/MI) and 30-day death using inverse probability weighting.

Results

The 30-day risk of death/MI was 14.8% (95% CI 11.8%-18.4%) for a <10g/dL strategy, 15.1% (95% CI 11.7%-18.2%) for a <9g/dL strategy, 15.9% (95% CI 12.4%-19.0%) for a <8g/dL strategy, and 18.3% (95% CI 14.6%-22.0%) for a <7g/dL strategy. Absolute risk differences and risk ratios, relative to the <10g/dL strategy, for 30-day death/MI increased as thresholds decreased, though 95% confidence intervals were wide. Findings were similar and imprecise for 30-day death.

Limitation

Unmeasured confounding may have persisted despite adjustment.

Conclusion

The 30-day risks of death/MI and death among patients with acute MI and anemia appear to rise progressively with lower hemoglobin concentration thresholds for transfusion. However, the imprecision around estimates from this target trial analysis preclude definitive conclusions about individual hemoglobin thresholds.

Primary Funding Source

National Heart, Lung, and Blood Institute: U01 HL133817, U01 HL132853.

INTRODUCTION

Recent results from the Myocardial Ischemia and Transfusion (MINT) trial suggested a restrictive transfusion strategy led to a 15% increased risk of 30-day all-cause death or recurrent myocardial infarction (MI) compared with a liberal strategy in 3,504 adults hospitalized with acute MI and anemia(1). In the design of the MINT trial, the two treatment strategies were based upon the precedent that most transfusion trials compared a liberal arm with hemoglobin thresholds between <9-10g/dL (<90-100g/L) and a restrictive arm with thresholds between <7-8g/dL (<70-80g/L)(2, 3). Additionally, the pragmatic design of the MINT trial permitted transfusion within a range of hemoglobin concentrations. As a consequence of these design choices, there was an opportunity to explore the effect of different hemoglobin thresholds for patients with acute MI and anemia.

The hemoglobin threshold at which red blood cell (RBC) transfusion affords greatest benefit, without increasing patient outcome risks, may be particularly important for patients with acute MI. Prior studies showed that risk of adverse outcomes increased progressively as anemia worsened in this clinical setting(4, 5). A recent review of transfusion trials noted insufficient evidence pertaining to treatment strategies opting for triggers between 8g/dL and 10g/dL in patients with acute MI(3). Current evidence alone cannot inform transfusion policies in this target population. Though conducting a randomized trial to study numerous transfusion thresholds would be ideal to inform clinical practice, the size and complexity of such a trial would be prohibitive.

Accordingly, we emulated a hypothetical target trial(6-14) and estimated the effect of four transfusion strategies defined by hemoglobin thresholds between 7g/dL and 10g/dL(15-22) on the composite of 30-day death or recurrent MI, leveraging data from MINT to derive additional insights on an optimal transfusion threshold for patients with acute MI. The results from this analysis complement MINT trial results to inform the optimal RBC transfusion threshold for this patient population.

METHODS

We pre-specified a secondary analysis of the MINT trial to evaluate effects of four different transfusion strategies (hemoglobin thresholds <10g/dL, <9g/dL, <8g/dL, and <7g/dL) to trigger RBC transfusion (in the absence of other clinical indications or contraindications for transfusion) on the composite risk of death or recurrent MI through 30 days among patients with acute MI and anemia (hemoglobin <10g/dL). Table S1 summarizes the hypothetical target trial protocol and emulation of this trial, as well as the protocol of the MINT trial.

The MINT trial was approved by the Institutional Review Boards at the coordinating centers, Rutgers University (Pro20160000722) and University of Pittsburgh (CR19040050). All sites had local approval.

Target Trial Design

Eligibility

The hypothetical target trial would include patients 18 years and older with type 1, 2, 4b, or 4c MI(23). For inclusion, patients require a hemoglobin concentration <10g/dL at randomization, measured during the index hospitalization. Other eligibility criteria would be consistent with the MINT protocol(24).

Randomized assignment

Eligible individuals would be randomly assigned, with equal probability, to one of four transfusion strategies.

Treatment strategies

The four transfusion strategies would specify that, unless otherwise indicated, an RBC transfusion is to be given within 24 hours each time a hemoglobin value less than X g/dL is obtained during hospitalization, where X takes the value of 10, 9, 8, or 7. Additionally, no transfusions would be given when the most recent hemoglobin level is at or above X g/dL. When indicated, blood is to be administered one unit of packed RBCs at a time, with each transfusion followed by a hemoglobin measurement within 24 hours. The assigned transfusion strategy should be maintained until hospital discharge, death, or 30 days post-randomization, whichever occurs first. Protocol exceptions to the randomized strategies allow for transfusions to treat adverse clinical events (e.g. bleeding, angina, surgery) and for transfusions to be withheld to avoid hazardous outcomes (e.g. fluid overload, cardiac arrest, transfusion reactions), regardless of the hemoglobin concentration (see Appendix A for details). In these circumstances, the transfusion protocol would be suspended until the clinical situation has been resolved.

Per guidance outlined in the MINT trial manual of operations, in the absence of clinical exceptions, violations to the target trial protocol would occur if: 1) transfusions are not given within 24 hours of a hemoglobin measurement below the specified threshold, 2) hemoglobin levels are not obtained within 24 hours following a transfusion, 3) transfusions are given above the specified threshold, 4) a participant is discharged within 30 days with a hemoglobin value below the assigned threshold, or 5) the transfusion strategy is not resumed within 36 hours following cardiac surgery (e.g., coronary artery bypass graft or transcatheter aortic valve replacement).

Outcomes

The primary outcome would be the composite of all-cause death or recurrent MI (death/MI). A secondary outcome is all-cause death. Participants would be followed until death, loss-to-follow-up/withdrawal, or day 30, whichever occurred first.

Causal Contrast

We would estimate the per-protocol effect of each hemoglobin transfusion threshold strategy. Therefore, we would evaluate the effect of adhering to each strategy during the index hospitalization for acute MI.

Analysis

Participants would be censored at the time they deviate from their assigned strategy or at last contact (if lost to follow-up or withdrew). As adherence is not randomized, we would adjust for selection bias introduced by censoring for nonadherence by assigning each patient a non-stabilized inverse probability (IP) weight at every time point they remain adherent(25-30). The denominator of these weights is, informally, the patient’s probability of adhering to the assigned strategy through time t, conditional on time-varying adverse events (e.g., coronary revascularization, bleed, heart failure, infection), current hemoglobin level, and baseline prognostic factors (i.e., MI type, renal function, revascularization for the index MI, any prior heart failure, bleeding, intubation, history of anemia, history of renal failure). We would estimate IP weights via a pooled logistic regression within each treatment arm.

To estimate the per-protocol effect of transfusion strategies on outcomes, we would fit a weighted pooled logistic regression, regressed on the assigned treatment, restricted cubic spline for time, and the following baseline covariates: age, hemoglobin concentration, intubation, bleeding, MI type, revascularization for the index MI, any prior heart failure, history of MI, history of anemia, history of cancer, history of atrial fibrillation, and history of renal failure. We would obtain treatment-specific risks at day 30 after standardizing the model estimates to the distribution of baseline covariates and compute 95% nonparametric bootstrap confidence intervals (CIs) from 1,000 samples(17, 31). Lastly, we would estimate Kaplan-Meier curves for cumulative incidence of outcomes.

Emulation of Target Trial

Data source and eligibility for the emulation

The MINT trial was a randomized, pragmatic trial designed to assess the effectiveness of restrictive (hemoglobin threshold <7-8g/dL) and liberal (hemoglobin threshold <10g/dL) transfusion strategies in patients with acute MI and anemia(24). MINT included 3,504 patients from 144 clinical sites in six countries(1). The primary outcome was a composite of 30-day death or MI. We used data from MINT to emulate and compare the target trial transfusion strategies with identical emulated eligibility as the target trial (Table S1).

Treatment strategies

We emulated the target trial transfusion strategies as follows. Time was measured in 4-hour intervals. A patient’s baseline hemoglobin was noted at start of follow-up. We denoted each time patients had a hemoglobin measurement in-hospital and designated the lowest concentration or concentration known at the time of transfusion at each time interval; we carried forward the last known concentration. The MINT trial protocol required hemoglobin measurements at least daily for the first three days post-randomization and were obtained at the discretion of clinicians thereafter; <5% of required measurements were missing.

We noted if a first transfusion was given at time t and if multiple transfusions were provided. MINT allowed clinicians to provide RBCs to patients in either arm at any time to treat acute bleeding events. Investigators provided the reason for every transfusion in both arms (e.g., hemoglobin indicated, angina, investigator-specified clinical indication), which allowed us to identify transfusions to treat adverse events and thus exceptions to the strategies.

Clinical teams reported the date, but not time, for the first episode of clinical diagnoses (e.g., bleed, heart failure, infection). These events were noted as time-varying confounders which may inform transfusion decisions.

Last, we identified and emulated days in which other allowable exceptions to the target trial occurred (Appendix A). MINT investigators reported dates, but not times, and clinical reasons for withholding transfusions when patients’ hemoglobin was <10g/dL in the liberal strategy (e.g., fluid overload, awaiting dialysis, specified clinical reason). Patients in the restrictive arm were not required to receive transfusions when hemoglobin was <8g/dL. Investigators also reported if/when, and the reason why, a patient was discontinued from their assigned strategy; clinically valid reasons constituted an approved exception to the target trial on the date of incidence. Last, dates and times of surgeries were reported, which informed suspension of the target trial protocol on the day of and 36-hour window following procedures.

Missing data on dates/times of hemoglobin measurements, transfusions, and time-varying events were minimal, and are noted in Appendix B.

Outcomes

We ascertained outcomes of 30-day death and MI from MINT data. MIs were adjudicated by a treatment-blinded central committee in MINT. The date, but not exact time, of outcomes was recorded; events were assigned to occur at 12PM on the date of occurrence. If transfusions or hemoglobin measurements occurred after 12PM on a date of death and no MI occurred on the date, the outcome was set at 6PM or 11PM depending on the latest transfusion or measure.

Follow-up period

Time zero was the hour of MINT randomization and follow-up ended at death, loss to follow-up/withdrawal, or 11PM on day 30, whichever occurred first.

Analysis

We estimated the per-protocol effect of the target trial emulation, which was analogous to the target trial analysis, as follows. To emulate randomization and improve statistical efficiency, we “assigned” patients to all four treatment strategies at start of follow-up by creating four copies (“clones”) of each patients’ data (Figure 1). Cloning was done in part because all MINT participants were consistent with all four treatment strategies at randomization. We sequentially determined compatibility of clones’ data with the assigned strategy at every time interval according to the rules outlined by the target trial. Once a clone was censored, lost to follow-up, or withdrew in a particular strategy, subsequent follow-up time in that strategy was disregarded.

Figure 1.

Examples of censoring for 12 clones (A7-A10, B7-B10, and C7-C10) of three MINT patients (Patients A, B, and C) under each of the 4 transfusion strategies, <7g/dL to <10g/dL

We demonstrate the cloning and censoring procedures for the per-protocol analysis of the target trial emulation using three patients A, B, and C and their hypothetical “clones” 7-10. We assume no protocol exceptions occurred for these patients during the days shown.

Patient A is censored from the <7g/dL and <8g/dL strategies for receiving a transfusion above these thresholds. Likewise, they are censored from the <9g/dL strategy for receiving a transfusion above this concentration at a later time. Their observed data is fully adherent to the <10g/dL strategy over the 3 days shown.

Patient B is censored from the <10g/dL strategy at the end of Day 1 because their hemoglobin was below this threshold for 24 hours without receiving an RBC transfusion. They are then censored from the <8g/dL and <7g/dL strategies on Day 2 for receiving an RBC transfusion above these thresholds. On Day 3, this patient experienced a death or MI while they were adherent to the <9g/dL strategy only; this outcome would only count toward following this strategy and their follow-up time would end there.

Patient C is censored from the <7g/dL strategy on Day 1 because they received a transfusion above this threshold. They experienced a death or MI on Day 2, so this outcome would count toward the <10g/dL, <9g/dL, and <8g/dL strategies. Their follow-up would end in all three strategies at this time.

Patients contribute person-hours to the analysis of each strategy until they are first censored or experience an outcome in that strategy.

We assessed baseline and time-varying characteristics of patients who remained adherent to each strategy after day one of follow-up and at hospital discharge or day 30, whichever occurred first. Such descriptions informed the differential contribution of patients with particular attributes to each strategy over time (i.e., potential confounding for adherence).

We conducted IP weighting as described in the target trial, except we fit two pooled logistic regressions in our original, unexpanded dataset to estimate patients’ probability of transfusion at each time point – one model was fit in the first day of follow-up and the other model was fit in all subsequent in-hospital follow-up time (see Appendix B for modeling details)(11, 17). We adjusted for covariates noted in the target trial, the assigned MINT treatment, and an indicator for if the current hemoglobin was <8g/dL given the MINT restrictive threshold; we included a product term between the MINT assignment and the indication for hemoglobin <8g/dL. In the second model, we also adjusted for adverse events that occurred on the day prior to time point t and whether patients were discontinued from their assigned MINT arm. Adherence probability was equal to one during any clinical exceptions, when hemoglobin was not remeasured following a transfusion, and any time greater than 24 hours since the last hemoglobin measurement. IP weights were defined as the inverse of the cumulative adherence probability at time t. We truncated weights at the 99^th percentile to mitigate near-positivity violations.

We carried out IP weighted analyses on 30-day death/MI as described in the target trial, but applied to the dataset of clones (see Appendix B for modeling details)(17, 31). We also adjusted for the assigned MINT arm in the outcome model. The absolute risks and 95% CIs in each strategy were calculated at days 10, 20, and 30 for Kaplan-Meier incidence curves. As a secondary analysis, the same analytic procedures were followed using death as the outcome.

Sensitivity Analyses

We fit unweighted pooled logistic regressions for the outcomes, including terms only for time, treatment strategy, and their interaction.

All analyses were conducted in SAS 9.1.4 (SAS Institute, Cary, NC, USA).

Role of the Funding Source

The MINT trial was funded by the National Heart, Lung, and Blood Institute (U01 HL133817, U01HL132853) and the Canadian Blood Services and Canadian Institutes of Health Research Institute of Circulatory and Respiratory Health (grant 342193). Supported, in part, by the RHU iVASC grant ‘#ANR-16-RHUS-00010’ from the French National Research Agency as part of the “Investissements d’Avenir” program. Funders had no role in the design, analysis, or reporting of this secondary analysis.

RESULTS

Of 3,504 MINT patients, 12 patients did not meet eligibility in the current analysis due to hemoglobin values above the required threshold (n=3), previous enrollment or enrollment in a competing study (n=3), or another/unknown violation to MINT trial eligibility criteria (n=6). Patients in this analysis (N=3,492) averaged age 72 years, 55% were male, and 79% of patients with known race were white (Table 1). The average hemoglobin concentration was 8.6g/dL at baseline and 56% of patients had a type 2 index MI.

Table 1.

Baseline characteristics of MINT patients eligible for target trial emulation (N=3,492)

Demographic Characteristics, n (%)	N=3,492
Age, mean (SD) (n=3,491)	72.1 (11.6)
Male (Sex)	1,907 (54.6)
Race (n=3,045)*
White	2,395 (78.7)
Black	417 (13.7)
Other	233 (7.7)
Smoking Status
Never	1,322 (37.9)
Former	1,425 (40.8)
Current	545 (15.6)
Unknown	200 (5.7)
Medical History, n (%)
Myocardial infarction	1,135 (32.5)
Anemia	1,488 (42.6)
Cancer	767 (22.0)
Renal failure/insufficiency	1,605 (46.0)
Stroke or TIA	615 (17.6)
Hypertension	2,968 (85.0)
Diabetes (treated with medication)	1,889 (54.1)
Index MI Characteristics/Patient Status, n (%)
MINT arm assignment
Restrictive	1,742 (49.9)
Liberal	1,750 (50.1)
Hemoglobin (g/dL), mean (SD)	8.6 (0.8)
MI Type†
ST-segment elevation MI	653 (18.7)
Type 1	1,454 (41.6)
Type 2	1,950 (55.8)
Active bleed	459 (13.1)
Intubation	481 (13.8)
History of heart failure, heart failure, or reduced LVEF	1,865 (53.4)
Renal function (n=3,482)
Dialysis prior to randomization	412 (11.8)
eGFR <30 mL/min/1.73m2	801 (23.0)
eGFR 30-60 mL/min/1.73m2	996 (28.6)
eGFR ≥60 mL/min/1.73m2	1,273 (36.6)
Revascularization for index MI	976 (28.6)

^*Race information not collected for European Union and Brazil sites

^†Not mutually exclusive

30-Day Death/MI Analysis

Adherence to the different target trial strategies for more than one day was associated with the MINT treatment assignment, discontinuation of assigned treatment, and post-randomization revascularization. Adherence at discharge/day 30 was associated with these same factors as well as baseline MI type, baseline renal function, and post-randomization infections (Table 2). The highest percentage of patients adhered to the <8g/dL strategy (35.3%), and the lowest percentage adhered to the <9g/dL strategy (19.0%) through 30 days (Table 3). The rate of transfusions decreased as the hemoglobin threshold decreased. In the two treatment models for IP weight estimation, the MINT treatment assignment and patients’ current hemoglobin level were strongly associated with transfusions; bleeding pre- and post-randomization were also associated with transfusions (Table S2). The distribution of IP weights during adherent follow-up time is shown in Table S3.

Table 2.

Characteristics of patients who remained adherent to target trial strategies for more than one day of follow-up and at hospital discharge in analysis of 30-day death or MI (N=3,492)

	Adherent for >1 Day				Adherent at Discharge or Day 30*
Characteristics, n (%)	<10g/dL Strategy	<9g/dL Strategy	<8g/dL Strategy	<7g/dL Strategy	<10g/dL Strategy	<9g/dL Strategy	<8g/dL Strategy	<7g/dL Strategy
Number adherent without outcome	1,851 (53.0%)	1,635 (46.8%)	1,825 (52.3%)	1,623 (46.5%)	1,050 (30.1%)	662 (19.0%)	1,231 (35.3%)	1,187 (34.0%)
Baseline
MI Type
Type 1	785 (42.4%)	699 (42.8%)	757 (41.5%)	673 (41.5%)	448 (42.7%)	309 (46.7%)	512 (41.6%)	487 (41.0%)
Type 2	1,030 (55.6%)	890 (54.4%)	1,015 (55.6%)	897 (55.3%)	575 (54.8%)	326 (49.2%)	677 (55.0%)	657 (55.3%)
Type 4b/4c/unknown	36 (1.9%)	46 (2.8%)	53 (2.9%)	53 (3.3%)	27 (2.6%)	27 (4.1%)	42 (3.4%)	43 (3.6%)
Renal Function
Unknown	8 (0.4%)	6 (0.4%)	3 (0.2%)	3 (0.2%)	5 (0.5%)	2 (0.3%)	2 (0.2%)	3 (0.3%)
Dialysis	210 (11.3%)	199 (12.2%)	233 (12.8%)	214 (13.2%)	94 (9.0%)	72 (10.9%)	141 (11.5%)	151 (12.7%)
eGFR <30 mL/min/1.73m2	427 (23.1%)	392 (24.0%)	432 (23.7%)	367 (22.6%)	213 (20.3%)	125 (18.9%)	241 (19.6%)	239 (20.1%)
eGFR 30-60 mL/min/1.73m2	522 (28.2%)	461 (28.2%)	516 (28.3%)	460 (28.3%)	320 (30.5%)	197 (29.8%)	360 (29.2%)	336 (28.3%)
eGFR ≥60 mL/min/1.73m2	684 (37.0%)	577 (35.3%)	641 (35.1%)	579 (35.7%)	418 (39.8%)	266 (40.2%)	487 (39.6%)	458 (38.6%)
Revascularization for index MI	512 (27.7%)	460 (28.1%)	522 (28.6%)	474 (29.2%)	306 (29.1%)	220 (33.2%)	386 (31.4%)	356 (30.0%)
Clinical bleed	243 (13.1%)	225 (13.8%)	251 (13.8%)	209 (12.9%)	123 (11.7%)	74 (11.2%)	163 (13.2%)	151 (12.7%)
Intubation	262 (14.2%)	220 (13.5%)	254 (13.9%)	218 (13.4%)	118 (11.2%)	70 (10.6%)	132 (10.7%)	132 (11.1%)
History of renal failure	855 (46.2%)	758 (46.4%)	845 (46.3%)	756 (46.6%)	457 (43.5%)	267 (40.3%)	532 (43.2%)	544 (45.8%)
History of anemia	793 (42.8%)	703 (43.0%)	776 (42.5%)	687 (42.3%)	446 (42.5%)	261 (39.4%)	518 (42.1%)	498 (42.0%)
MINT Assignment
Liberal	1,448 (78.2%)	585 (35.8%)	246 (13.5%)	164 (10.1%)	976 (93.0%)	234 (35.3%)	122 (9.9%)	98 (8.3%)
Restrictive	403 (21.8%)	1,050 (64.2%)	1,579 (86.5%)	1,459 (89.9%)	74 (7.0%)	428 (64.7%)	1,109 (90.1%)	1,089 (91.7%)
Time-Varying †
Current hemoglobin (g/dL)‡	9.9 (9.0, 1.5)	9.2 (8.6, 9.8)	8.8 (8.3, 9.4)	8.8 (8.3, 9.4)	10.8 (10.4, 11.3)	9.9 (9.4, 10.5)	9.1 (8.6, 9.8)	9.0 (8.5, 9.7)
Discontinued from Liberal assignment	16 (0.9%)	35 (2.1%)	47 (2.6%)	49 (3.0%)	22 (2.1%)	52 (7.9%)	50 (4.1%)	45 (3.8%)
Discontinued from Restrictive assignment	3 (0.2%)	3 (0.2%)	3 (0.2%)	2 (0.1%)	2 (0.2%)	6 (0.9%)	20 (1.6%)	21 (1.8%)
Bleed	51 (2.8%)	43 (2.6%)	38 (2.1%)	25 (1.5%)	77 (7.3%)	38 (5.7%)	89 (7.2%)	75 (6.3%)
Heart Failure	74 (4.0%)	57 (3.5%)	51 (2.8%)	40 (2.5%)	120 (11.4%)	61 (9.2%)	119 (9.7%)	118 (9.9%)
Embolism	7 (0.4%)	7 (0.4%)	6 (0.3%)	4 (0.2%)	18 (1.7%)	6 (0.9%)	17 (1.4%)	19 (1.6%)
Bacteremia or pneumonia	43 (2.3%)	42 (2.6%)	44 (2.4%)	35 (2.2%)	56 (5.3%)	22 (3.3%)	75 (6.1%)	74 (6.2%)
Suspected MI§	9 (0.5%)	8 (0.5%)	10 (0.5%)	8 (0.5%)	8 (0.8%)	5 (0.8%)	9 (0.7%)	9 (0.8%)
Revascularization	205 (11.1%)	155 (9.5%)	161 (8.8%)	140 (8.6%)	261 (24.9%)	132 (19.9%)	266 (21.6%)	249 (21.0%)
Stroke or TIA	8 (0.4%)	9 (0.6%)	7 (0.4%)	4 (0.2%)	12 (1.1%)	7 (1.1%)	16 (1.3%)	15 (1.3%)
Unstable Angina	8 (0.4%)	10 (0.6%)	8 (0.4%)	6 (0.4%)	31 (3.0%)	22 (3.3%)	42 (3.4%)	35 (2.9%)
Surgery	6 (0.3%)	5 (0.3%)	4 (0.2%)	4 (0.2%)	20 (1.9%)	10 (1.5%)	20 (1.6%)	21 (1.8%)

^*At day 30 if still in-hospital and adherent

^†Occurred on any date prior to noted day during hospitalization

^‡Median (Q1, Q3) of known hemoglobin values at end of denoted time period

^§Reported by clinical staff as MI in-hospital, may or may not have been adjudicated by central committee

Table 3.

Risks of 30-day death or MI and death for RBC transfusion thresholds of <7g/dL to <10g/dL (N=3,492)

Treatment Strategy	<10g/dL Strategy	<9g/dL Strategy	<8g/dL Strategy	<7g/dL Strategy
30-Day Death or MI
Adherence and Events
No. of person-hours	901,128	594,876	1,005,100	951,924
No. of person-hours in-hospital	276,292	183,116	268,716	249,536
No. transfusions	3,810	1,872	1,238	621
No. transfusions per 100,000 hours	422.8	314.7	123.2	46.6
Complete 30-day adherence	1,050 (30.1%)	662 (19.0%)	1,231 (35.3%)	1,187 (34.0%)
Outcome Estimates
No. of outcomes	200	154	241	230
30-day absolute risk (%) (95% CI)	14.8% (11.8, 18.4)	15.1% (11.7, 18.2)	15.9% (12.4, 19.0)	18.3% (14.6, 22.0)
30-day risk difference (%) (95% CI)	Reference	0.3% (−3.7, 4.0)	1.0% (−4.5, 6.3)	3.5% (−2.5, 9.5)
30-day risk ratio (95% CI)	Reference	1.02 (0.78, 1.33)	1.07 (0.74, 1.51)	1.24 (0.86, 1.78)
30-Day Death
Adherence and Events
No. of person-hours	940,188	624,384	1,051,732	997,992
No. of person-hours in-hospital	310,244	203,204	305,072	281,504
No. transfusions	3,982	1,892	1,280	434
No. transfusions per 100,000 hours	423.5	303.0	121.7	43.5
Complete adherence	1,117 (32.0%)	717 (20.5%)	1,312 (37.6%)	1,270 (36.4%)
Outcome Estimates
No. of outcomes	94	61	130	130
30-day absolute risk (%) (95% CI)	7.8% (5.3, 10.7)	7.8% (5.1, 10.5)	11.3% (7.8, 14.4)	14.0% (9.7, 17.7)
30-day risk difference (%) (95% CI)	Reference	0.1% (−3.3, 3.3)	3.5% (−2.2, 7.9)	6.2% (−0.4, 11.3)
30-day risk ratio (95% CI)	Reference	1.01 (0.62, 1.53)	1.45 (0.80, 2.37)	1.80 (0.96, 2.94)

The adjusted estimated 30-day risk of death/MI ranged from 14.8% (95% CI 11.8%-18.4%) in the <10g/dL strategy to 18.3% (95% CI 14.6%-22.0%) in the <7g/dL strategy (Table 3). Relative to the <10g/dL strategy, the absolute risk differences and risk ratios increased as the hemoglobin thresholds decreased, though there was minimal difference between the <10g/dL and <9g/dL strategies. All 95% CIs included the null values. An unweighted outcome model estimated effects of lower magnitude and showed a similar trend (Table S4). The IP weighted Kaplan-Meier curves for time to death/MI are shown in Figure 2A.

Figure 2.

Inverse-probability weighted, estimated cumulative incidence curves for (A) time to death/MI and (B) time to death for four RBC transfusion strategies over 30-days

30-Day Death Analysis

Associations between patient characteristics and adherence to the target trial strategies over time were similar in the death analysis to those observed in the death/MI analysis (Table S5). The patterns for duration of follow-up and adherence by target trial strategy (Table 3) and factors associated with transfusions (Table S6) were also comparable to what was observed in the death/MI analysis. The distribution of IP weights is in Table S3.

The adjusted estimated 30-day risk of death was highest for the <7g/dL strategy (14.0% [95% CI 9.7%-17.7%]) (Table 3). Relative to the <10g/dL threshold, the <8g/dL and <7g/dL strategies suggested increased risk of death, while there was no difference with the <9g/dL strategy; 95% CIs were imprecise. The unadjusted 30-day estimates showed a similar pattern (Table S4). The weighted Kaplan-Meier curves are presented in Figure 2B.

DISCUSSION

This target trial emulation of 3,492 patients with acute MI and anemia included in the MINT trial suggested risk of 30-day death/MI and death was highest for hemoglobin transfusion thresholds <8g/dL and <7g/dL, but not different for a <9g/dL threshold, all relative to a <10g/dL threshold. However, 95% CIs for the effect estimates were wide and included null values.

Two previous randomized trials reported estimates for restrictive versus liberal RBC transfusion strategies in patients with acute MI and anemia. The MINT trial (N=3,504) found a restrictive transfusion strategy led to a considerable, yet non-significant, increase in incidence of 30-day death/MI and 30-day death compared to a liberal strategy(1). The Restrictive and Liberal Transfusion Strategies in Patients with Acute MI trial (N=668) estimated a non-inferior rate of 30-day major adverse cardiovascular events from a restrictive versus liberal strategy(32). Two pilot trials also reported conflicting evidence(33, 34). Observational studies demonstrated inverse relationships between admission hemoglobin and adjusted risk of death in patients with acute coronary syndrome, suggesting more severe anemia is associated with adverse outcomes(4, 5, 35-37). No prior studies have reported effects of distinct hemoglobin thresholds for transfusion on outcomes.

Similar to MINT trial results, this analysis estimated a higher incidence of 30-day death/MI and death from lower transfusion thresholds compared to a <10g/dL threshold in this population of patients with acute MI; the present results included wide 95% CIs. The IP weighted Kaplan-Meier curves further suggested increased incidence of outcomes from <8g/dL and <7g/dL thresholds over 30 days compared to the <9g/dL and <10g/dL thresholds. Prior evidence indicated a restrictive transfusion strategy is safe and effective in most clinical settings, with the exception of patients with acute coronary syndrome; there has previously been a lack of evidence(3). Estimates from the current analysis are in line with pathophysiologic evidence indicating potential harms of anemia in the presence of acute coronary syndrome(38-40).

The present per-protocol analysis complements MINT trial intention-to-treat results and may contribute to guiding clinical practice in patients with acute MI(25). The present findings suggest transfusion guidelines could consider avoiding hemoglobin triggers of <8g/dL and <7g/dL in this patient population to minimize risk of hazardous outcomes, recognizing that imprecision in the estimates precludes definitive conclusions about the individual thresholds. A <9g/dL strategy showed minimal difference in effect from a <10g/dL strategy on both outcomes. A target of 9g/dL will reduce blood use and avoid potential risks associated with additional exposure to RBCs, such as heart failure or thrombogenicity(1, 41). Though inferences from this analysis involve uncertainty, other studies have provided little guidance about whether clinicians should use a <9g/dL or <10g/dL threshold to treat anemia in patients with acute MI.

The causal interpretation of our estimates relied on “successful” emulation of the target trial protocol elements. There were some limitations to our analysis. First, in the MINT trial, hemoglobin measurements were not required daily following the first three days post-randomization. This may have limited data for our emulation. Second, reporting of protocol violations in the MINT trial was differential between treatment arms, which may have biased our ability to capture adherence to target trial strategies; investigators reported reasons why transfusions were withheld in the liberal arm, but not in restrictive. However, transfusions were not required when patients’ hemoglobin was <8g/dL in the restrictive arm, which would violate this target trial strategy regardless. Third, unmeasured confounding related to time-varying events may persist. For example, we did not have information on blood supplies and staffing shortages at clinical sites, which may have impacted adherence to a target trial strategy. Fourth, time-varying confounders were measured by dates, not time, and only the first incidence of each type of clinical event was reported. Repeated events may have been reported as protocol violations or clinically valid reasons for interruption of MINT strategies. As a result, we may have misclassified patients as not experiencing a recurrent event or as having an event earlier on a date than actually occurred. Unmeasured confounding or model misspecification may have resulted(42, 43). However, baseline covariates were well measured in MINT and missing dates/times for time-varying factors was minimal. The limitation in obtaining times also applied to outcomes, though outcomes were well measured given adjudication procedures for MI and 98.5% follow-up completion in our sample. Last, we were limited to assessing transfusion thresholds between the pre-defined range 7-10g/dL; it is unknown if thresholds above 10g/dL improve outcomes for this patient population.

Despite the limitations, this target trial emulation provided effect estimates for transfusion strategies that have not been previously reported from randomized trials (i.e., separate effects for <8g/dL and <7g/dL strategies). We estimated absolute and relative risks to facilitate clinical interpretation and simulated CIs to demonstrate the degree of uncertainty in our analyses. This study may have been underpowered to detect the estimated effect sizes, potentially leading to inexact estimates. However, this pre-specified analysis was based on randomized clinical trial data, which may have improved reliability and specificity in data reporting, making this a unique target trial analysis to increase potential knowledge gained from a single clinical trial(44, 45). Specific MINT trial eligibility criteria may limit generalizability if characteristics of the included patients are not representative of the larger target population.

In conclusion, relative to a <10g/dL RBC transfusion strategy, this target trial emulation estimated that the 30-day risks of death/MI and death increased progressively with lower hemoglobin concentration thresholds for transfusion signaling that more restrictive thresholds may increase incidence of death or MI in patients with acute MI. The imprecision of these estimates, however, prevents explicit recommendations regarding individual hemoglobin transfusion thresholds.