Artifactual Declines in Durable LVAD Utilization among Heart Transplant Candidates: Selection Bias in Transplant Registries

Abstract

Background:

The Organ Procurement and Transplantation Network (OPTN) Standard Transplant Analysis and Research (STAR) files are commonly used to study use of mechanical circulatory support (MCS) among heart transplant candidates. However, the STAR MCS file reports devices only for candidates removed from the waitlist, whereas the Scientific Registry of Transplant Recipients (SRTR) MCS file includes devices for all candidates. We compared the STAR and SRTR MCS files in estimating trends in LVAD support and waitlist outcomes among heart transplant candidates with durable LVADs.

Methods:

We identified adult heart transplant candidates who obtained durable LVADs between October 18, 2018 and December 31, 2024 using MCS files from STAR (N=4,182) and SRTR (N=4,919). We estimated the cumulative incidence of transplantation and death/deterioration in both cohorts and analyzed LVAD prevalence at bimonthly intervals from October 2018 to April 2025.

Results:

The SRTR cohort showed stable LVAD prevalence (30.7%−37.4%). In contrast, the STAR cohort constructed from the MCS file showed a 92.5% decline in LVAD prevalence (1,740 to 131 candidates), despite 971 candidates listed with status 4 for LVAD support on April 30, 2025. The STAR cohort demonstrated higher 3-year cumulative incidence of transplantation (66.7% vs 54.4%) and death/deterioration (10.1% vs 4.5%) after LVAD implantation. When we reran STAR-based analyses with both MCS and status justification files, prevalence and outcomes matched SRTR estimates.

Conclusions:

The STAR MCS file excludes LVADs among currently listed candidates, producing artifactual declines in LVAD prevalence and incidence and inflated waitlist outcomes. Awareness of registry construction is essential for accurate interpretation of LVAD-related outcomes.

X Handle:

“Two registries, same data, wildly different durable LVAD counts. New paper from @aanieldhn et al. explains why OPTN STAR undercounts LVAD prevalence by >90% and how to fix it.”

INTRODUCTION

After the adult donor heart allocation policy change in October 2018, which deprioritized stable LVAD-supported candidates from status 1B to status 4, recent reports have shown that rates of durable left ventricular assist device (LVAD) support on the heart transplant waitlist have declined precipitously¹. Given the clinical benefits of LVADs, including significant improvements in functional status^2,3 and equivalent 2-year survival compared with heart transplant^4,5, understanding temporal trends and optimizing allocation policies for LVAD-supported candidates remain high priorities for heart allocation policymakers. Thus, accurate data on LVAD-related trends and waitlist outcomes are essential to inform these decisions.

Much of the literature reporting trends and outcomes among LVAD-supported heart transplant candidates has utilized the Standard Transplant Analysis and Research (STAR) files provided by the Organ Procurement and Transplantation Network (OPTN)^1,6. The STAR dataset’s MCS file contains information about devices only among candidates who have been removed from the waitlist, systematically excluding those among currently listed candidates. In contrast, the Scientific Registry of Transplant Recipients (SRTR), which receives identical raw data from the OPTN, constructs its MCS file by incorporating device information from status justification forms, thereby including currently listed as well as transplanted and delisted candidates.

Researchers commonly rely on the default MCS files provided, leading to potential systematic bias when using the STAR dataset. We hypothesized that the reported declining LVAD trends in literature utilizing STAR data may be artifactual due to the STAR MCS file’s exclusion of currently listed candidates. We evaluated whether reliance on removal-based device data in the STAR MCS file significantly alters estimates of LVAD-related trends and waitlist outcomes compared to complete device data that includes currently listed candidates.

METHODS

Data Sources

This study used data from the OPTN STAR files and the SRTR Standard Analysis Files (SAF). All OPTN members are required to submit comprehensive data on all transplant candidates, recipients, and donors. The STAR files represent publicly available, deidentified patient-level extracts of the OPTN database. The SRTR, operating under federal contract to support the OPTN, receives a complete copy of the OPTN database and applies additional data supplementation processes before releasing its own SAF, leading to differences in some outcomes such as long-term post-kidney transplant survival⁷. The datasets are available upon request from the SRTR and OPTN.

Heart transplant programs submit MCS device information to the OPTN through multiple data collection instruments. At initial listing, centers report device use on Transplant Candidate Registration (TCR) forms, and for transplant recipients, on Transplant Recipient Registration (TRR) forms. Between initial listing and waitlist removal, there are two distinct instruments for capturing ongoing MCS use: 1) status justification forms, submitted throughout active listing to document current device support and 2) removal forms, submitted only when candidates are removed from the waitlist. Both instruments exist in both the STAR and SRTR databases. The critical difference is which instrument each registry uses to construct its primary MCS file. STAR's MCS file (“thoracic_mcs_device”) uses only removal data (capturing devices at waitlist removal). Waitlist removal in the STAR’s MCS file includes all removal reasons, such as transplantation, death, clinical deterioration, clinical recovery, etc. In contrast, the SRTR prepares its MCS file (“ThorSupportDevice”) by collating status justification forms, capturing devices for all candidates including those still listed. This structural choice, rather than differential data availability, creates the systematic bias documented in this study. The absence of status justification forms in the STAR’s MCS file is not mentioned in the Data Dictionary. Furthermore, while the OPTN STAR User Guide reports this device file as containing data “collected at waitlist removal,” it does not emphasize that devices among currently listed candidates are excluded. Copies of this guide and the data dictionary have been made available on the GitHub repository for this manuscript⁸. This study was deemed exempt by the Stanford University Institutional Review Board.

Study Objectives

The primary objectives were to 1) compare identification and prevalence of durable LVAD support between the STAR and SRTR MCS files, 2) assess temporal trends in LVAD prevalence and incidence on the waitlist, and 3) evaluate differences in waitlist outcomes (transplantation versus death/deterioration).

Study Population

We identified all adult (>17 years of age at initial listing) heart transplant candidates who obtained durable LVADs between October 18, 2018 and December 31, 2024 using the MCS files from both the STAR and SRTR registries. We collected relevant demographic and clinical data from patients at the time of LVAD implantation. Variables included sex, age at LVAD implantation, body mass index (BMI), race and ethnicity, blood type, primary diagnosis, and insurance type. We identified all implanted durable LVADs using the primary MCS files in both databases. As detailed in the Supplemental Methods, we did not account for LVAD explantation in any of our analyses. Given published explantation rates of <1–2%^6,9,10, we reasoned that not accounting for explantation has negligible impact on our findings.

Trends in LVAD Prevalence and Incidence on the Waitlist

To assess LVAD prevalence over time according to each registry’s MCS file, we calculated the number of candidates with active LVAD support at bimonthly intervals from October 18, 2018 through April 30, 2025. At each timepoint, a candidate was considered to have active LVAD support if their LVAD implantation date preceded the assessment date and they had not been removed from the waitlist. The denominator was all listed candidates regardless of inclusion in the registry’s MCS file. We compared prevalence estimates between the STAR and SRTR datasets at each timepoint.

We also examined the composition of status 4 candidates—the lowest priority status for stable candidates with durable LVADs—at the same bimonthly timepoints. At each timepoint, we linked status 4 candidates to status justification files to identify which candidates had status 4 specifically for durable LVAD support rather than other criteria, such as inotropic support without invasive hemodynamic monitoring.

We then replicated the work of Mullan et al.¹ and assessed temporal trends in the incidence of LVAD support by analyzing waitlist flow dynamics in consecutive 2-month periods from January 2019 through April 2025. For each period, we quantified (1) new listings with LVAD—candidates listed during the period whose LVAD implantation date preceded their listing date and (2) LVAD implants during waitlisting—candidates whose LVADs were implanted after their initial listing date. We compared these flow metrics between the STAR and SRTR datasets to identify systematic differences in capturing LVAD prevalence and incidence patterns.

Competing Risks Analysis

We assessed the cumulative incidence of waitlist outcomes following durable LVAD implantation between October 18, 2018 and December 31, 2024 until waitlist removal with administrative censoring on June 30, 2025. We treated death or waitlist removal for clinical deterioration and transplantation as competing risks and estimated cumulative incidence curves separately for the SRTR and STAR cohorts. Candidates removed from the waitlist for other reasons, including clinical recovery and voluntary withdrawal, were censored at the time of removal. To quantify the magnitude of selection bias, we calculated the absolute difference in cumulative incidence between STAR and SRTR post-LVAD implantation up to 3 years. Because the STAR MCS file captures only candidates who have been removed from the waitlist, the STAR cohort is inherently conditioned on reaching a definitive outcome.

For candidates who received LVADs prior to listing, defining LVAD implantation date as time zero precedes the period during which waitlist outcomes could occur, introducing potential immortal time bias. Thus, we performed a sensitivity analysis in which time zero was defined as the later of LVAD implantation date or initial listing date, ensuring all patients were both listed and LVAD-supported at the start of follow-up.

Reconstitution of Complete LVAD Cohort Using Status Justification Files in STAR

To determine whether the OPTN STAR systematic undercount could be corrected to match SRTR estimates, we tested whether combining the MCS file with status justification forms would reconstitute the complete LVAD cohort and yield equivalent prevalence trends and outcomes. We constructed a reconstituted OPTN STAR cohort by identifying all patients with LVAD support from both the MCS device file and all status justification forms with documented LVAD support. We then repeated all prevalence and outcomes analyses using this reconstituted cohort to assess whether it reproduced SRTR results.

Generalizability to Other Mechanical Circulatory Support Devices

To assess whether the systematic exclusion of currently listed candidates extends beyond durable LVADs, we performed an analogous prevalence analysis for intra-aortic balloon pumps (IABPs). We identified adult heart transplant candidates supported by IABPs using device type codes in both the SRTR device file and STAR MCS file. We calculated the prevalence of IABP support among actively listed candidates at bimonthly intervals from January 2020 through January 2025 using the same methodology as the LVAD prevalence analysis.

Statistical Analysis

The cohorts constructed from both STAR and SRTR capture overlapping populations of the same patients. As a result, similar to prior work¹¹, we did not perform any statistical testing to determine whether there were any differences between cohorts as this would assume independent populations. Instead, we compared the demographics, clinical characteristics, and outcomes descriptively. We performed all analyses with R (version 4.5.2). Complete statistical code to reproduce the study results is available online⁸.

RESULTS

Cohort Identification and Baseline Characteristics

Between October 18, 2018 and December 31, 2024, we identified 4,919 adult heart transplant candidates with durable LVAD support in the SRTR cohort and 4,182 in the STAR cohort. Table 1 shows the baseline characteristics of candidates at the time of LVAD implantation; there were no missing data. Median age at listing (SRTR: 55 years [IQR 46–62] vs STAR: 56 years [IQR 46–62]) and at LVAD implantation (SRTR: 54 years [IQR 44–60] vs STAR: 55 years [IQR 45–61]) were similar. The SRTR cohort had a higher proportion of HeartMate 3 devices (86.7% vs 81.8%) while a higher percentage of LVADs in the STAR cohort were older models (HeartMate II [2.6% vs 1.0%] and HVAD [14.6% vs 11.8%]). Median follow-up in the SRTR cohort was longer at 1.88 years (IQR 1.05–3.12 years) than in the STAR cohort at 1.42 years (IQR 0.70–2.49 years).

Table 1.

Baseline Characteristics of LVAD Cohorts by Data Source

Characteristic	SRTR N = 4,919	OPTN STAR N = 4,182
Age at Listing (Years)	55 (46, 62)	56 (46, 62)
Age at LVAD Implant (Years)	54 (44, 60)	55 (45, 61)
Sex
Male	3,937 (80.0%)	3,227 (77.2%)
Female	982 (20.0%)	955 (22.8%)
Race/Ethnicity
White	2,679 (54.5%)	2,347 (56.1%)
Black	1,542 (31.3%)	1,270 (30.4%)
Hispanic/Latino	405 (8.2%)	392 (9.4%)
Asian	114 (2.3%)	100 (2.4%)
Other	179 (3.6%)	73 (1.7%)
BMI Category
Normal (18.5–24.9 kg/m2)	719 (14.6%)	779 (18.6%)
Underweight (<18.5 kg/m2)	16 (0.3%)	21 (0.5%)
Overweight (25–29.9 kg/m2)	1,593 (32.4%)	1,420 (34.0%)
Obese (≥30 kg/m2)	2,576 (52.4%)	1,958 (46.8%)
Unknown	15 (0.3%)	4 (0.1%)
Blood Type
A	1,651 (33.6%)	1,483 (35.5%)
AB	161 (3.3%)	154 (3.7%)
B	627 (12.7%)	579 (13.8%)
O	2,480 (50.4%)	1,966 (47.0%)
Primary Diagnosis
Dilated Cardiomyopathy, Non-Ischemic	2,380 (48.4%)	2,056 (49.2%)
Ischemic Cardiomyopathy	1,587 (32.3%)	1,374 (32.9%)
Restrictive Cardiomyopathy	801 (16.3%)	585 (14.0%)
Congenital Heart Disease	26 (0.5%)	20 (0.5%)
Other	125 (2.5%)	147 (3.5%)
Insurance Type
Private	2,024 (41.1%)	1,695 (40.5%)
Public	2,807 (57.1%)	2,454 (58.7%)
Other	88 (1.8%)	33 (0.8%)
LVAD Device Type
HeartMate 3	4,265 (86.7%)	3,421 (81.8%)
HeartMate II	51 (1.0%)	109 (2.6%)
HVAD	582 (11.8%)	612 (14.6%)
Other	21 (0.4%)	40 (1.0%)

LVAD Prevalence on the Waitlist

We observed systematic underreporting in STAR’s MCS file compared to SRTR’s MCS file across all bimonthly intervals from October 2018 to April 2025. The total number of status 4 candidates was similar over time in both datasets (Figure 1, Tables S1 and S2). On October 18, 2018, both databases showed similar LVAD prevalence (SRTR MCS: 32.8%, STAR MCS: 34.8%) (Figure S1). However, while SRTR prevalence remained relatively stable throughout the study period (30.7%−37.4%), STAR showed a dramatic decline, falling to 3.2% by April 30, 2025. According to STAR’s MCS file, there were 1,740 patients with durable LVADs on October 18, 2018 and only 131 on April 30, 2025 even though 971 people were actively listed with status 4 justifications for durable LVAD support on that date according to the status justification forms (Figure 1).

Figure 1: Trends in LVAD Prevalence among Heart Transplant Candidates.

Stacked bars show total candidates with status 4 justifications (blue) and the subset with status 4 justifications specifically for LVAD support (orange) at bimonthly intervals between October 18, 2018 and April 30, 2025. The red line shows the total candidates with LVADs recorded in the MCS device file across all statuses. Using SRTR data (bottom panel), LVAD prevalence per the MCS file remains stable throughout the study period. In OPTN STAR data (top panel), LVAD prevalence per the MCS file declines from 1,740 on October 18, 2018 to 131 on April 30, 2025 despite 971 candidates having status 4 justifications for LVAD support on that date. Note: The red line reflects total LVAD prevalence across waitlist statuses and is not limited to status 4 candidates represented by the bars.

Temporal Trends in the Incidence of LVADs on the Waitlist

In early 2019, both STAR and SRTR showed similar absolute counts with approximately 170–190 candidates listed with pre-existing LVADs every 2 months (Figure 2, Tables S3 and S4). By late 2024, SRTR captured 130–160 candidates per bimonthly period while STAR captured only 70–90. The proportion of new listings with pre-existing LVAD support declined from 15.8% (Jan-Feb 2019) to 9.8% (Mar-Apr 2025) in SRTR while in STAR this proportion declined more dramatically from 15.2% to 1.5% over the same period. Both databases demonstrated similarly low rates of candidates who had LVADs implanted after initial listing throughout the study period.

Figure 2: Temporal Trends in Incidence of LVAD Utilization on the Heart Transplant Waitlist.

Solid lines represent candidates added to the waitlist at bimonthly intervals with pre-existing LVAD support from January 2019 to April 2025. Dashed lines represent candidates who had LVADs implanted after initial listing.

Competing Risks Analysis

Cumulative incidence rates of transplant and death/deterioration were substantially higher in the cohort constructed with the STAR MCS file. For transplantation, relative to the SRTR cohort, STAR overestimated the cumulative incidence at 3 years after LVAD implantation at 66.7% (95% CI [65.1%, 68.2%]) vs 54.4% (95% CI [52.9%, 55.9%]) (Figure 3). Similarly, for death/deterioration, STAR’s 3-year cumulative incidence was more than 6 percentage points higher at 10.1% (95% CI [9.2%, 11.1%]) vs 4.5% (95% CI [3.9%, 5.2%]).

Figure 3: Cumulative Incidence of Transplantation and Death/Deterioration among LVAD-Supported Candidates.

The cohort constructed using the STAR MCS file overestimates both competing outcomes compared to that using the SRTR MCS file. At 3 years post-LVAD implantation, STAR overestimates transplantation by 12.3 percentage points (66.7% vs 54.4%) and death/deterioration by 5.6 percentage points (10.1% vs 4.5%). Numbers at risk and cumulative events are shown below the curves.

In a sensitivity analysis using the later of LVAD implantation or listing date as time zero, the discrepancy in 3-year cumulative incidence between SRTR and STAR persisted: (death/deterioration: 7.3% [95% CI 6.4%, 8.2%] vs. 12.6% [95% CI 11.5%, 13.7%]; transplantation: 68.2% [95% CI 66.5%, 69.8%] vs. 77.9% [95% CI 76.4%, 79.3%]) (Figure S2).

Reconstitution of the OPTN STAR LVAD Cohort Using Status Justification Files

When we combined the STAR MCS file with all LVAD records from candidate registration and status justification files, the reconstituted cohort included 4,980 candidates compared to 4,182 in the STAR MCS file alone and 4,919 in the SRTR MCS file. Absolute counts in incidence of LVAD support and waitlist outcomes were nearly identical between the reconstituted STAR cohort and the cohort constructed from the SRTR MCS file (Figures S3 and S4). At 3-years, the cumulative incidence of transplantation was 54.9% (95% CI [53.4%, 56.4%]) in the STAR cohort versus 54.4% (95% CI [52.9%, 55.9%]) in the SRTR. The cumulative incidence of death/deterioration at 3 years was 4.5% (95% CI [3.9%, 5.2%]) in both cohorts. Furthermore, the prevalence of LVAD support in the reconstituted STAR cohort aligned almost perfectly with SRTR estimates (Figure 4).

Figure 4: Trends in A) Absolute Number of Candidates and B) Proportion of the Waitlist Supported by Durable LVADs.

Sole use of the MCS file in STAR (blue) shows a precipitous decrease in LVAD prevalence over time, down to only 131 candidates on April 30, 2025. However, supplementation of the MCS file with status justification files in the STAR registry to include LVADs of currently listed candidates yields prevalence estimates (green) that are nearly identical to those from the SRTR MCS file (orange). The orange and green lines in Panel A are overlapping.

Generalizability to Intra-Aortic Balloon Pumps

From January 2020 to January 2025, IABP prevalence according to the STAR MCS file was consistently lower than in the SRTR MCS file (Table S5), with absolute differences ranging from 18 to 88 candidates across all bimonthly timepoints. However, by combining the STAR MCS file with all status justifications indicating IABP support produced estimates almost identical to SRTR (Figure S5).

DISCUSSION

In this study, we found that sole use of the MCS file in the OPTN STAR dataset yields 1) trends in LVAD support that decline artifactually over time and 2) substantial overestimations in the cumulative incidence of waitlist mortality and transplantation. This major discrepancy occurs because the STAR MCS file captures devices only among candidates who have been removed from the waitlist. Linking status justification files with the MCS file in STAR produces a complete cohort with outcomes and prevalence/incidence trends that match those estimated from the SRTR MCS file. Furthermore, this selection bias affects not just durable LVADs but other devices as well, including IABPs, with important implications for published research on MCS-supported heart transplant candidates.

Multiple lines of evidence confirmed this selection mechanism. First, the STAR cohort had substantially shorter median follow-up time (1.42 vs 1.88 years) because it captures only those who reach definitive outcomes and excludes stable patients still actively listed with long-standing device support. Additionally, baseline characteristics differed predictably, with STAR having higher proportions of older LVAD models expected in patients implanted earlier who have since been delisted. Furthermore, by the end of the study period, the number of candidates with status 4 justifications for durable LVAD support far exceeded the total number of LVADs captured in STAR’s MCS file, a discrepancy that is necessarily due to systematic under-capture of currently listed LVAD-supported candidates. Addition of status justification files to the STAR MCS file reconstituted identical waitlist outcome and prevalence trends as the SRTR MCS file.

The impact of this systematic bias on published literature about candidates supported by LVADs is likely substantial given the frequent use of the OPTN STAR files^1,6,12–16. Using OPTN STAR data, Mullan et al. previously reported that the incidence of LVAD support among heart transplant candidates declined precipitously after the 2018 policy change, with new listings with LVADs approaching almost zero by mid-2020¹. However, using the SRTR MCS file and the STAR MCS file supplemented with status justification files, we identified a substantial number of new waitlist additions of patients with LVADs as recently as 2025. Bridge-to-transplant LVAD implantations during listing have decreased, consistent with current literature^12,17. However, a stable bimonthly volume of LVAD-supported patients continues to newly enter the waitlist with devices already in place rather than through the traditional bridge-to-transplant pathway.

The selection bias we documented also distorts waitlist outcomes substantially. Researchers who solely use the STAR MCS file to identify LVAD cohorts inherently select for patients reaching definitive outcomes while excluding stable patients who are still listed with ongoing device support, inflating estimates of both transplantation and death/deterioration. Because registry-derived outcome estimates inform clinical expectations, benchmarking, and policy evaluation, ensuring their validity is essential for both clinical interpretation and allocation system design. This is particularly consequential given the OPTN Heart Committee recent and ongoing focus on determining appropriate waitlist prioritization of candidates supported by durable LVADs¹⁸.

While our analysis focused on durable LVADs, this structural limitation affects all MCS devices. We showed that prevalence of IABP support on the waitlist is underestimated when relying solely on STAR’s MCS file. Researchers may reasonably assume that the primary MCS device file provides comprehensive device information because it is the default file provided in the STAR registry. However, without explicit documentation of its limitations, researchers unfamiliar with the nuances of OPTN data collection may inadvertently introduce systematic bias by relying on this file alone. Our results highlight how registry file structure can introduce bias in longitudinal analyses even when registries utilize identical raw data, underscoring the importance of transparency in registry documentation and careful data linkage practices.

Recommendations

We offer three recommendations. First, the OPTN should update its STAR Data Dictionary and User Guide to explicitly state that the MCS file captures devices only at waitlist removal and excludes all currently listed candidates. Given how frequently researchers utilize STAR to study MCS among heart transplant candidates, prominently documenting this limitation would substantially reduce the risk of inadvertent bias. Alternatively, the OPTN could restructure the STAR MCS file to include device information for all listed candidates, mirroring the approach already implemented by the SRTR. Second, researchers using OPTN STAR to study MCS trends or outcomes should supplement the MCS file with status justification forms and device fields from TCR and TRR registration forms; complete linkage code is available in our public repository⁸. Third, researchers should fully disclose not just the data source used but the specific files, variables, and linkages employed; journal editors and reviewers should require this level of transparency for all transplant registry analyses.

Limitations

We did not account for LVAD explantation among current listed candidates due to data limitations. Both SRTR and reconstituted STAR prevalence estimates may therefore overestimate the true number of candidates with functioning LVADs. However, given published explantation rates of <1–2% and that there were between 1,200 and 1,800 candidates supported by LVADs at any given timepoint between October 2018 and April 2025, potential overestimation would be at most 30–40 candidates—a negligible impact on our core findings. Second, we did not perform sex- and gender-stratified analyses, as the primary aim of this study was to characterize registry-level methodological differences rather than clinical outcome disparities.

Conclusions

Durable LVAD use in heart transplant candidates has been previously reported to decrease significantly after the 2018 heart allocation policy change. Our findings indicate that this narrative may reflect structural under-ascertainment of currently listed LVAD candidates in commonly used OPTN STAR files rather than true reductions in LVAD support among candidates awaiting transplantation. Supplementation of the STAR MCS file with status justification forms yields estimates equivalent to the SRTR, revealing stable LVAD prevalence and incidence among listed candidates. Because OPTN data directly inform solid organ allocation policy in the United States, awareness of how transplant registry files are constructed is essential for accurate inference and policy-relevant conclusions.

Abbreviations:

OPTN

Organ Procurement and Transplantation Network

SRTR

Scientific Registry of Transplant Recipients

STAR

Standard Transplant Analysis and Research

SAF

Standard Analysis Files

LVAD

Left ventricular assist device

MCS

Mechanical circulatory support

IABP

Intra-aortic balloon pump

BMI

Body mass index

CI

Confidence interval