PreOpNet showed limited performance with AUC 0.70 for death and 0.67 for MACE, overestimated risk, and had lower discrimination than hs-cTnT in European non-cardiac surgery patients.
Does the PreOpNet deep learning algorithm improve preoperative risk assessment for 30-day death and MACE compared to hs-cTnT and RCRI in adults undergoing noncardiac surgery?
The PreOpNet deep learning algorithm overpredicts 30-day mortality risk and has lower discriminative ability than hs-cTnT for preoperative risk assessment in non-cardiac surgery.
Absolute Event Rate: 0% vs 0%
Abstract Introduction PreOpNet is a novel deep-learning algorithm for the preoperative risk assessment of all-cause death and major adverse cardiac events (MACE) within 30 days post-operatively. Its performance in a European clinical setting and against high-sensitivity cardiac troponin (hs-cTnT), recommended by the latest European Society of Cardiology guidelines for pre-operative risk assessment in patients undergoing non-cardiac surgery, remains unknown. Methods In this secondary analysis of a single-centre diagnostic cohort study, consecutive adult patients undergoing noncardiac surgery with a planned postoperative stay of ≥ 24 hours were prospectively enrolled from October 2014 to September 2019. Patients were excluded if a preoperative 12-lead ECG was missing. All patients received a systematic screening using serial measurements of hs-cTnT in clinical routine. Model performance was assessed using the area under the receiving operating characteristic curve (AUC) and calibration curves and compared to the revised cardiac risk index (RCRI) and preoperative hs-cTnT. The primary outcome was 30 days post-procedural all-cause death. The secondary outcome was 30 days MACE. Findings Among 6,106 patients with an available 12-lead ECG, (median age 74.0 p25-p75: 69.0, 80.0, 2657 (43.5%) female), 219 (3.6%) patients died and 434 (7.1%) had a MACE within 30 days following the procedure. Model performance was limited, with a low-to-moderate AUC for death and MACE (0.70 95% CI: 0.67-0.74 and 0.67 95% CI: 0.65-0.70, respectively) and a strong overestimation of risk (calibration intercept: -2.00 -2.14 – -1.86 and -2.35 -2.45 – -2.25; calibration slope: 0.71 0.58-0.84 and 0.61 0.51-0.71, respectively). PreOpNet had lower discrimination than hs-cTnT for both death and MACE (difference in AUC 0.05 95% CI: 0.01-0.10 and 0.07 95% CI: 0.04-0.10, respectively; pdeath=0.02, pmace0.001). In contrast, PreOpNet outperformed RCRI for death (difference in AUC 0.06 95% CI: 0.01-0.11; p=0.02) but not for MACE (difference in AUC 0.01 95% CI: -0.04-0.03; p=0. 70). Interpretation The performance of PreOpNet decreased when validated in a prospective cohort recruiting consecutive adult patients undergoing non-cardiac surgery in a European clinical setting. PreOpNet overpredicted the 30-day mortality risk and demonstrated lower discrimination than hs-cTnT, making the algorithm unlikely to offer additional value to clinicians for risk stratification or in guiding decisions of proceeding with non-cardiac surgeries.Comparison of discriminative performance Calibration curves of the PreOpNet
Champetier et al. (Sat,) reported a other. PreOpNet showed limited performance with AUC 0.70 for death and 0.67 for MACE, overestimated risk, and had lower discrimination than hs-cTnT in European non-cardiac surgery patients.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: