Table 2.
Performance of the local, pooled, and federated LASSOa and MLPb models at each site, based on AUROCsc with 95% confidence intervals.
| Model |
Mount Sinai Brooklyn (n=611), AUROC (95% CI) | Mount Sinai Hospital (n=1644), AUROC (95% CI) | Mount Sinai Morningside (n=749), AUROC (95% CI) | Mount Sinai Queens (n=540), AUROC (95% CI) | Mount Sinai West (n=485), AUROC (95% CI) | |
| LASSO model | ||||||
| Local | 0.791 (0.788-0.795) | 0.693 (0.689-0.696) | 0.66 (0.656-0.664) | 0.706 (0.702-0.710) | 0.482 (0.473-0.491) | |
| Pooled | 0.816 (0.814-0.819) | 0.791 (0.788-0.794) | 0.789 (0.785-0.792) | 0.734 (0.730-0.737) | 0.829 (0.824-0.834) | |
| Federated | 0.793 (0.790-0.796) | 0.772 (0.769-0.774) | 0.767 (0.764-0.771) | 0.694 (0.690-0.698) | 0.801 (0.796-0.807) | |
| MLP model | ||||||
| Local | 0.822 (0.820-0.825) | 0.750 (0.747-0.754) | 0.747 (0.743-0.751) | 0.791 (0.788 - 0.795) | 0.719 (0.711-0.727) | |
| Pooled | 0.823 (0.820-0.826) | 0.792 (0.789-0.795) | 0.751 (0.747-0.755) | 0.783 (0.779-0.786) | 0.842 (0.837-0.847) | |
| Federated (no noise | 0.829 (0.826-0.832) | 0.786 (0.782-0.789) | 0.791 (0.788-0.795) | 0.809 (0.806-0.812) | 0.836 (0.83-0.841) | |
aLASSO: least absolute shrinkage and selection operator.
bMLP: multilayer perceptron.
cAUROC: area under the receiver operating characteristic curve.