Richard E. Clark Memorial Paper for Perioperative and Critical Care: The Society of Thoracic Surgeons Intermacs National Database Risk Model for Durable Left Ventricular Assist Device Implantation

F. D.. Pagani¹, B. Singletary², R. Cantor³, A. Nayak⁴, j. teuteberg⁵, J. Mehaffey⁶, p. shah⁷, J. cowger⁸, D. Vega⁹, D. Goldstein¹⁰, P. Kurlansky¹¹, J. Stehlik¹², T. Dardas¹³, J. Jacobs¹⁴, D. Shahian¹⁵, R. Habib¹³, J. Kirklin^3
1University of Michigan Hospital, Ann Arbor, Michigan ²KIRSO, birmingham, Alabama ³University of Alabama in Birmingham, Birmingham, Alabama ⁴Baylor Scott and White Dallas, Dallas, Texas ⁵Stanford University, Palo Alto, California ⁶West Virginia University, Morgantown, West Virginia ⁷Inova Fairfax Hospital, Fairfax, Virginia ⁸Henry Ford Medical Center, Detroit, Michigan ⁹Emory University, Atlanta, Georgia ¹⁰Montefiore Medical Center, Bronx, New York ¹¹Division of Cardiothoracic Surgery, Columbia University Irving Medical Center, New York, New York ¹²University of Utah, Salt Lake City, Utah ¹³The Society of Thoracic Surgeons, CHICAGO, Illinois ¹⁴University of Florida, Gainesville, Florida ¹⁵Massachusetts General Hospital, Sudbury, Massachusetts

Purpose: Risk models for durable left ventricular assist device (dLVAD) implantation inform candidate selection and facilitate quality improvement. Prior risk models were based on clinical trial populations. We developed a risk model for 90-day mortality utilizing a real-world experience from The Society of Thoracic Surgeons National Intermacs Database (STS Intermacs).

Methods: STS Intermacs was queried for primary implants of fully magnetically levitated continuous flow (FML) dLVADs implanted from January 2019 through September 2023. Exclusion criteria included isolated or concurrent right-sided device (n=913), total artificial heart (n=83), previous dVAD (n=149), or missing follow-up information (n=34). The primary outcome was 90-day mortality. Multivariable logistic regression was used to derive a risk model based upon pre-implant risk factors using a derivation cohort (DC; 2019-2021). Final variables were retained using 1000 sampled models (stepwise selection p< 0.05, >40% models). The model was validated using 2022-2023 dVAD implants as a validation cohort (VC). Model performance was assessed in both the DC and VC cohorts using area under the curve (AUC or C statistic), Brier Scores, and calibration plots. A refined model based on all patients was generated and used to calculate the observed to expected [O/E with 95% confidence intervals (CI)] ratios aggregate for each program.

Results: The study population consisted of 11,342 patients who received a primary isolated FML dLVAD at 176 centers participating in STS Intermacs from 2019-2023. The study cohort was sequentially divided in time into a derivation cohort (DC; n=6,775) and validation cohort (VC; n=4,567). Intermacs Profile 1 or 2 (47.2% vs. 52.2%, p< 0.0001) and concomitant surgery (49.3% vs. 55.5%, p< 0.0001) were more likely in VC, while prior coronary bypass surgery was more likely in DC (15.0% vs. 12.3%, p < 0.0001). Overall, 90-day mortality was 8.0% (9.2% in DC vs 7.4% in VC, p=0.001). Logistic regression risk model derived for 90-day mortality applied to both the DC and the VC produced similar discrimination (AUC .71 for both) and calibration (Brier score .08 vs .07), with overestimation of risk among patients with predicted risk > 0.4. The risk model further refined to all 11,342 implants was similarly discriminating (AUC = 0.70) and calibrated (0.07). The corresponding expected risk estimates were used to derive O/E ratios for all 176 centers. O/E analysis identified 22 (12.5%) centers with higher-than-expected mortality (O/E [95% CI] > 1) and 14 (8.0%) lower-than-expected mortality centers (O/E [95% CI] < 1) (all p<.05) (Figure).

Conclusion: The STS Intermacs Risk Model demonstrated satisfactory performance and calibration. Few high-risk candidates limited model performance at extreme risk, suggesting that current clinical practice carefully considers expectations for high-risk candidates. The STS Intermacs Risk Model represents an important prognostic tool to inform candidate selection and facilitate quality improvement.

Identify the source of the funding for this research project: The data for this research were provided by The Society of Thoracic Surgeons’ National Database. Data analysis was performed at the STS Intermacs Data Coordinating Center.