Author + information
- Jennifer A. Cowger, MD, MS∗ ()
- ↵∗Reprint requests and correspondence:
Dr. Jennifer Cowger, Department of Cardiovascular Medicine, St. Vincent Heart Center of Indiana 8333 Naab Road, Suite 400, Indianapolis, Indiana 46260.
Risk prediction is integral to the field of advanced heart failure, both for clinical decision making and for the purposes of discussing risk for informed patient consent at the time of mechanical circulatory support (MCS) implantation or transplantation. However, in this young field of MCS, the accuracy and precision of mortality risk prediction in MCS patients face several challenges. MCS is presently in a state of continued and rapid evolution. Over the past 5 years, the field has witnessed a marked increase in the number of device implants, and substantial gains have been made in patient survival, the latter attributed to advancements in technology, refinement of surgical techniques, and improvements in patient selection (1). Mortality risk prediction during MCS is therefore also an evolving process, and new correlates of risk will continually be identified while (simultaneously) previously known predictors of MCS mortality may change with regard to the magnitude of their attributable risk. For patients who are supported on MCS as a bridge to transplant (BTT), the complexity of risk prediction is amplified. In addition to the need to predict survival on MCS, survival after transplantation, inclusive of both donor and recipient factors, must be taken into account.
In this issue of JACC: Heart Failure, Johnston et al. (2) attempted to help elucidate the conundrum of risk prediction in BTT MCS patients by examining outcomes in 6,036 transplant patients registered in the United Network for Organ Sharing (UNOS) database who were previously on ventricular assist device support. They developed and internally validated a risk score, called the TRIP-MCS (Transplantation Risk Index in Patients With Mechanical Circulatory Support), using 1-year mortality risk correlates identified from heart transplant recipients supported by MCS and their donors. Approximately 70% of transplant recipients supported by MCS were status 1A at the time of transplantation, but it is unclear whether these patients were using their allowance of 30 days of elective status 1A time or whether they were classified as status 1A for urgent device complication. Recipient risk correlates included older recipient age, higher body mass index, need for pre-transplantation mechanical ventilation or intensive care unit stay, renal/hepatic dysfunction, presence of an infection within 2 weeks of transplantation, and type of ventricular assist device support used before transplantation. Donor risk correlates of post-transplantation mortality included known risk factors such as older donor age, donor-recipient sex mismatch, longer ischemic time, and lower donor glomerular filtration rate. In both the derivation and validation groups, each 1-point increase in the TRIP-MCS increased the odds of post-transplantation mortality by 8%. The authors suggest that ∼46% to 49% of TRIP-MCS high-risk BTT patients could move to the moderate-risk group with improved donor selection.
There are 5 important attributes of a model to consider when the quality of the TRIP-MCS is assessed (3,4). 1) Is the model derived from a sample representative of the population of interest? 2) Is the model derived by use of standardized covariates with sound statistics? 3) Is the model accurate (calibration and discrimination)? 4) Has the model been shown to work (i.e., has it been validated)? 5) Will the model provide useful information with an impact on clinical decision making? The TRIP-MCS model was derived with UNOS data and is therefore largely representative of heart transplant patients bridged with MCS within the United States. The TRIP-MCS may not be applicable beyond U.S. borders, where listing rules/statuses, donor-recipient selection, average ischemic time, and post-MCS and transplantation management may vary substantially. The risk correlates that constitute the TRIP-MCS were clearly defined according to UNOS, and outcomes measured were per standard definition; however, UNOS data (and therefore the TRIP-MCS) do not take into account critical covariates that may impact patient outcome on continued MCS or outcomes after cardiac transplantation. Variables not accounted for in patients who continue on MCS could include active or prior device hemolysis or thrombosis, recurrent bouts of gastrointestinal bleeding, significant right ventricular failure, the added morbidity and mortality from complication-driven device exchange, and the increased hazard of death and complications with longer-term MCS support (1). Data are not tallied on those BTT MCS patients who die before transplantation. Furthermore, it is unclear how many patients underwent transplantation as a 1A listing status for MCS complications versus 1A elective time. Although the authors discuss selecting an “alternative organ” to improve BTT MCS recipient risk, the urgency with which many device complications present and the limited donor pool may make the option to “wait” unfeasible for several transplant candidates.
In terms of model accuracy and validation (3,4), the TRIP-MCS performs modestly well. The model was internally validated with a separate UNOS patient sample, and the overall model area under the curve was 0.66, which suggests modest model discrimination exists. Model calibration, defined as the agreement between model-predicted and observed mortality, was good for scores <20. The TRIP-MCS score can range from 0 to 60, but the mean score was 14.7 ± 7. Patients in the low-risk group (TRIP-MCS score ≤10) had predicted and observed mortality rates of 8.6% and 7.1%, respectively, whereas high-risk patients (score >20) had predicted and observed mortality rates of 31% and 26.7%, respectively.
The last criterion for a successful model is one of the hardest attributes to prove in the field of advanced heart failure: Will the model provide useful information and change one’s pre-test probability for patient outcome? Age and MCS type are nonmodifiable risk factors at the time of transplantation. Certainly, some stable MCS patients listed as status 1B or with 1A elective time may benefit from waiting for a more optimal donor match. In lower-risk BTT patients, the TRIP-MCS may have a beneficial role during physician/patient shared decision making and discussions about using 1A elective time on MCS. However, some stable MCS BTT patients, such as larger individuals with an O blood type or those with high panel reactive antibodies, may have a small pool of “perfect” donors, and waiting may not be to their benefit. The presence of high-risk TRIP-MCS covariates (such as the need for pre-transplantation mechanical ventilation, intensive care unit stay, and significant renal/hepatic dysfunction) likely signifies a patient status that is heading toward death on MCS or toward “bail-out” high-risk transplantation. In this setting, one could argue the recipient may be too sick to allow for judicious organ allocation; a lower-risk recipient should receive the organ. Conversely, is a model area under the curve of 0.66 good enough to comfortably make this call without taking into account the other unmeasured factors? The famous statistician George Box once said, “Essentially, all models are wrong, but some are useful” (5). The utility of the TRIP-MCS can only be known once it has been applied prospectively to patients and the impact of delayed transplantation on wait-list mortality and post-transplantation survival has been scrutinized.
↵∗ Editorials published in JACC: Heart Failure reflect the views of the authors and do not necessarily represent the views of JACC: Heart Failure or the American College of Cardiology.
Dr. Cowger has reported that she has no relationships relevant to the contents of this paper to disclose.
- American College of Cardiology Foundation
- Johnston L.E.,
- Grimm J.C.,
- Magruder J.T.,
- Shah A.S.
- Moons K.G.,
- Kengne A.P.,
- Woodward M.,
- et al.
- DeGroot M.H.