Statistical risk models for durable left ventricular assist device (LVAD) implantation inform candidate selection, quality improvement, and evaluation of provider performance. We developed a 90-day mortality risk model utilizing The Society of Thoracic Surgeons National Intermacs Database (STS Intermacs). STS Intermacs was queried for primary durable LVAD implants from 1/ 2019 - 9/2023. Multivariable logistic regression was used to derive a model based upon pre-implant risk factors using derivation (2019-2021 implants) and validation (2022-2023 implants) cohorts. Model performance (derivation and validation cohorts) was assessed using C-statistics, Brier Scores, and calibration plots. A refined model (all patients) was generated to calculate observed/expected [O/E, 95% confidence intervals (CI)] ratios for each center. The study population consisted of 11,342 patients from 2019-2023 sequentially divided in time into derivation (n=6,775) and validation (n=4,567). Ninety-day mortality was 8.0% (9.2% in derivation cohort vs. 7.4% in validation cohort; p=0.001). Logistic regression applied to derivation and validation cohorts produced similar discrimination (area under the curve (AUC) 0.714, CI: 0.69-0.74 and AUC 0.707, CI: 0.67-0.72, respectively) and calibration (Brier score .08 vs .07), with overestimation of risk among patients with predicted risk > 0.4. The O/E analysis identified 22 (12.5%) centers with worse-than-expected mortality with a CI > 1.0 and 14 centers (8.0%) with better-than-expected mortality with a CI < 1.0 (all p < 0.05). The STS Intermacs Risk Model demonstrated satisfactory discrimination and calibration. This tool may be used to inform candidate selection, facilitate quality improvement, and assess provider performance.
Read full abstract