Highlights
-
•
Adherence to the assigned transfusion strategy was higher in the restrictive than the liberal transfusion strategy group.
-
•
Patients who adhered to the assigned transfusion strategy tended to be lower risk than patients who did not adhere.
-
•
In a range of per-protocol analyses, patients receiving a liberal transfusion strategy had lower rates of death or MI and death than patients receiving a restrictive transfusion strategy.
-
•
Instrumental variable analysis results were consistent with the intention-to-treat findings.
ABSTRACT
We conducted a series of analyses to estimate the per-protocol effect of restrictive versus liberal transfusion strategies on 30-day death and death or recurrent myocardial infarction (MI) among patients from the Myocardial Ischemia and Transfusion (MINT) trial. Multiple analytic approaches were used to estimate the per-protocol effect, including analyses based on different definitions of adherence (site-reported, time-based, transfusion delay) and an instrumental variable analysis. Analyses based on adherence definitions found a restrictive strategy was associated with a substantially greater risk of 30-day death or recurrent MI, while the instrumental variable analysis estimated a moderate treatment effect similar to the original ITT result.
Background
The Myocardial Ischemia and Transfusion (MINT) trial randomized 3504 patients with myocardial infarction (MI) and anemia (hemoglobin <10 g/dL) to either a restrictive or liberal transfusion strategy. Using an intention-to-treat (ITT) approach, the primary manuscript reported a nonsignificant increased risk of 30-day death or recurrent MI for the restrictive compared with a liberal transfusion strategy (relative risk [RR]: 1.15 [95% CI: 0.99-1.34)]). This primary finding approached statistical significance ( P =.07) and suggested potential harm from the restrictive strategy. For the secondary outcome of 30-day death, a similar nonsignificant increase in risk was identified with the restrictive compared to liberal strategy (RR: 1.19 [95% CI: 0.96-1.47).
While the ITT approach provides a generalizable and unbiased estimate of the assigned treatment effect, it may underestimate the received treatment effect among participants who adhered to their randomized strategy. Per-protocol analyses, in contrast, are designed to estimate a treatment’s effect under ideal conditions (i.e., full adherence), supporting patient and provider decision-making by offering insight into what happens when a treatment is taken as intended. Given the borderline significant ITT result of the primary MINT analysis, it is critical to understand if the moderate treatment effect was a true mechanistic finding or if a larger, clinically significant per-protocol effect was attenuated by nonadherence. This manuscript reports results from a series of per-protocol analyses of the MINT trial data to determine the effect of receiving each transfusion strategy on risk of death or death or MI under full adherence to the trial protocol.
Methods
With the MINT restrictive strategy, transfusions were permitted only when the hemoglobin was <8 g/dL or if patients experienced persistent anginal symptoms. In contrast, the liberal strategy mandated transfusions to achieve a hemoglobin level ≥10 g/dL within 24 hours of randomization and maintain it at that level thereafter. The assigned protocol was followed throughout the index hospitalization or for 30 days, whichever came first. To facilitate these per-protocol analyses, sites reported whether the assigned transfusion strategy was maintained for each participant and recorded all hemoglobin measurements and red blood cell transfusions.
We used multiple per-protocol analytic approaches ( Table 1 ). First, we conducted an adjusted ‘adherer-only’ analysis. Multivariable regression models were performed to obtain adjusted treatment risk ratios (RR) specifically among those classified as adherent to their assigned strategy. To build the adjustment model for this analysis, missing baseline data were imputed, and univariable logistic regression was used to identify baseline variables associated with site-reported adherence ( p <.20). Key factors selected for adjustment included age, hemoglobin, country, renal function, history of peripheral artery disease, prior heart failure, symptoms of ischemia, new ST-T changes, and MI diagnosis confirmed at discharge. When possible, ventilator use and intensive care unit status at randomization were also included in the model due to their strong associations with study outcomes.
Table 1
Intention-to-treat, per-protocol, and instrumental variable definitions.
| Method | Patient population and treatment adherence definition | Analysis approach |
|---|---|---|
| Intention-to-treat (ITT) | All randomized participants ( N = 1,749 Restrictive; N = 1,755 Liberal) and assigned treatment. | Unadjusted analysis of assigned treatment on outcome. |
| Site-reported Maintenance of Protocol | Eligible participants with site-reported maintenance of assigned strategy or who discontinued strategy due to adverse events; those who discontinued due to patient or provider preference were excluded (i.e., considered non-adherent). | Multivariable log-binomial regression analysis of treatment received on outcome adjusting for factors associated with discontinuation of assigned treatment. |
| Proportion of Time Adherent | Eligible participants with ≥80%, or ≥70%, of their person-hours adherent to assigned transfusion protocol (calculated hourly) during index hospitalization. Allow clinical exceptions as defined by protocol. | Multivariable adjusted log-binomial regression analysis of treatment received on outcome. |
| Maximum Allowable Transfusion Delay | Eligible liberal patients received transfusions within 24-hours, or 36-hours, of each measured hemoglobin <10 g/dL. Eligible restrictive patients did not receive transfusion when hemoglobin >8 g/dL. Allow clinical exceptions as defined by protocol. | Multivariable adjusted log-binomial regression analysis of treatment received on outcome. |
| Instrumental Variable Analysis of Treatment Initiation | Eligible randomized participants ( N = 1,742 Restrictive; N = 1,750 Liberal). Randomized assignment is the instrumental variable, and treatment strategy is defined as receiving a transfusion (Liberal) or not receiving a transfusion (Restrictive) within 24 hours of randomization (i.e., treatment received is the exposure). Allow clinical exceptions as defined by protocol for the first 24 hours. | Instrumental variable analysis using 2-stage residual inclusion to estimate effect of treatment initiation on outcome. |
Second, we performed an instrumental variable analysis to estimate the per-protocol effect of initiating a restrictive as compared to a liberal transfusion strategy where initiation of the liberal strategy was defined as delivery of at least one red blood cell transfusion within 24 hours. Standard instrumental variable analyses assume a single, time-invariant treatment. Because the MINT transfusion strategy is a sustained treatment composed of a sequence of decisions over time, this instrumental variable analysis is limited to evaluating the causal effect of initiating treatment without considering whether the treatment was maintained.
Results
Overall adherence to the restrictive strategy was high across all per-protocol definitions (97.4%-99.8%), whereas adherence to the liberal strategy was lower and varied by definition (70.2%-86.7%) ( Figure ). Baseline characteristics of adherent and nonadherent participants are shown in Table 2 . In the restrictive strategy, nonadherent participants were clinically sicker at baseline; for example, they had lower left ventricular ejection fraction, worse renal function, were more likely to have a history of peripheral artery disease and be in the intensive care unit at randomization ( Table 2 ). In the liberal strategy, nonadherent participants had slightly lower baseline hemoglobin and were significantly less likely to have their initial MI diagnosis confirmed at discharge ( Table 2 ). Both groups also differed by enrollment location.
Adherence by various definitions and the effect of treatment on trial outcomes . Because of model non-convergence, prior ventilator use and ICU status at randomization were not included as covariates in the ‘time adherent’ death models, and prior ventilator use, ICU status, and symptoms of ischemia were excluded for the site-reported maintenance and maximum delay death models.
Table 2
Baseline characteristics of adherent participants using the site-reported definition.
| Liberal | Restrictive | ||||
|---|---|---|---|---|---|
| Variable | Level |
Nonadherent
N = 230 |
Adherent
N = 1,518 |
Nonadherent
N = 45 |
Adherent
N = 1,696 |
| Age (years), median (Q1, Q3) | 73.0 (63.9, 81.7) | 72.9 (65.1, 80.0) | 76.9 (69.7, 81.3) | 73.5 (65.0, 80.3) | |
| Sex, n (%) | Female | 95 (41.3%) | 720 (47.4%) | 23 (51.1%) | 746 (44.0%) |
| Enrollment location, n (%) | United States | 154 (67.0%) | 922 (60.7%) | 24 (53.3%) | 1,051 (62.0%) |
| Canada | 46 (20.0%) | 398 (26.2%) | 7 (15.6%) | 432 (25.5%) | |
| South America | 1 (0.4%) | 51 (3.4%) | 3 (6.7%) | 48 (2.8%) | |
| New Zealand/ Australia | 2 (0.9%) | 15 (1.0%) | 0 (0%) | 16 (0.9%) | |
| France | 27 (11.7%) | 132 (8.7%) | 11 (24.4%) | 149 (8.8%) | |
| Race available 1 , n (%) | 202 (87.8%) | 1,335 (87.9%) | 31 (68.9%) | 1,499 (88.4%) | |
| Race 2 , n (%) | White | 156 (77.2%) | 1,058 (79.3%) | 26 (83.9%) | 1,168 (77.9%) |
| Black | 24 (11.9%) | 187 (14.0%) | 5 (16.1%) | 207 (13.8%) | |
| Other | 22 (10.9%) | 90 (6.7%) | 0 (0%) | 124 (8.3%) | |
| Tobacco smoker, n (%) | Never | 89 (38.7%) | 621 (40.9%) | 18 (40.0%) | 680 (40.1%) |
| Former | 102 (44.3%) | 651 (42.9%) | 22 (48.9%) | 733 (43.2%) | |
| Current | 39 (17.0%) | 246 (16.2%) | 5 (11.1%) | 283 (16.7%) | |
| Medical history | |||||
| MI, n (%) | 69 (30.0%) | 477 (31.4%) | 12 (26.7%) | 575 (33.9%) | |
| Percutaneous coronary intervention, n (%) | 75 (32.6%) | 500 (32.9%) | 15 (33.3%) | 606 (35.7%) | |
| Coronary artery bypass graft, n (%) | 44 (19.1%) | 343 (22.6%) | 7 (15.6%) | 363 (21.4%) | |
| Heart failure, n (%) | 83 (36.1%) | 456 (30.0%) | 17 (37.8%) | 509 (30.0%) | |
| Number of vessels with >50% obstruction, n (%) | None | 6 (2.6%) | 55 (3.6%) | 1 (2.2%) | 60 (3.5%) |
| One | 20 (8.7%) | 203 (13.4%) | 7 (15.6%) | 223 (13.1%) | |
| Two | 32 (13.9%) | 177 (11.7%) | 5 (11.1%) | 218 (12.9%) | |
| Three | 42 (18.3%) | 285 (18.8%) | 8 (17.8%) | 330 (19.5%) | |
| Missing | 130 (56.5%) | 798 (52.6%) | 24 (53.3%) | 865 (51.0%) | |
| LV ejection fraction available, n (%) | 162 (70.4%) | 1,109 (73.1%) | 28 (62.2%) | 1,248 (73.6%) | |
| Most recent LV ejection fraction (%) within the past year, mean (SD) | 45.3 (35.0, 57.0) | 50.0 (38.0, 59.5) | 45.0 (35.5, 50.0) | 50.0 (38.9, 57.8) | |
| Stroke or transient ischemic attack, n (%) | 48 (20.9%) | 252 (16.6%) | 7 (15.6%) | 308 (18.2%) | |
| Atrial fibrillation, n (%) | 57 (24.8%) | 387 (25.5%) | 12 (26.7%) | 439 (25.9%) | |
| Peripheral artery disease, n (%) | 53 (23.0%) | 315 (20.8%) | 16 (35.6%) | 332 (19.6%) | |
| eGFR, n (%) | <30 | 55 (23.9%) | 341 (22.5%) | 7 (15.6%) | 398 (23.5%) |
| 30-59 | 58 (25.2%) | 436 (28.7%) | 13 (28.9%) | 491 (29.0%) | |
| ≥60 | 78 (33.9%) | 569 (37.5%) | 13 (28.9%) | 617 (36.4%) | |
| On dialysis at baseline | 39 (17.0%) | 172 (11.3%) | 12 (26.7%) | 190 (11.2%) | |
| Diabetes, n (%) | 124 (53.9%) | 820 (54.0%) | 23 (51.1%) | 920 (54.2%) | |
| Cancer, n (%) | 50 (21.7%) | 320 (21.1%) | 11 (24.4%) | 385 (22.7%) | |
| Anemia, n (%) | Acute | 75 (32.6%) | 570 (37.5%) | 15 (33.3%) | 627 (37.0%) |
| Chronic | 155 (67.4%) | 948 (62.5%) | 30 (66.7%) | 1,069 (63.0%) | |
| Index hospitalization, pre-randomization | |||||
| Type of MI, n (%) | Type 1 | 89 (38.7%) | 638 (42.0%) | 20 (44.4%) | 707 (41.7%) |
| Type 2 | 136 (59.1%) | 848 (55.9%) | 24 (53.3%) | 939 (55.4%) | |
| Other/ Unknown | 5 (2.2%) | 32 (2.1%) | 1 (2.2%) | 50 (2.9%) | |
| Percutaneous coronary intervention, n (%) | 68 (29.6%) | 450 (29.6%) | 12 (26.7%) | 523 (30.8%) | |
| Heart failure, n (%) | 64 (27.8%) | 338 (22.3%) | 12 (26.7%) | 364 (21.5%) | |
| Intubated on ventilator, n (%) | 34 (14.8%) | 197 (13.0%) | 10 (22.2%) | 240 (14.2%) | |
| Active bleeding, n (%) | 21 (9.1%) | 192 (12.6%) | 8 (17.8%) | 238 (14.0%) | |
| Diagnosis of MI at time of randomization confirmed at time of discharge, n (%) | 211 (91.7%) | 1,461 (96.2%) | 45 (100.0%) | 1,616 (95.3%) | |
| In ICU/CCU at randomization, n (%) | 106 (46.1%) | 719 (47.4%) | 31 (68.9%) | 817 (48.2%) | |
| Laboratory Values | |||||
| Baseline hemoglobin, g/dL, median (Q1, Q3) | 8.6 (7.9, 9.1) | 8.7 (8.1, 9.3) | 8.7 (8.0, 9.3) | 8.7 (8.0, 9.3) | |
| Most recent creatinine prior to randomization, mg/dL, median (Q1, Q3) | 1.4 (1.0, 3.2) | 1.4 (0.9, 2.5) | 1.4 (1.1, 3.0) | 1.4 (0.9, 2.5) | |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
