tavi-backend / artifacts /model_comparison.json
Jainish Solanki
Initial backend Space
84da040
Raw
History Blame Contribute Delete
11.4 kB
{
"n_train": 4000,
"n_test": 1000,
"base_rate": 0.0476,
"cohort": "synthetic_v1_seed42_n5000",
"n_bootstraps": 500,
"brdav_on_test": {
"name": "brdav_followup",
"label": "Br\u00fcggemann 2024 (real cohort, n=1,449) \u2014 different endpoint",
"calibration_curve": [
{
"decile": 0,
"n": 100,
"pred_mean": 0.21244,
"actual_mortality": 0.01
},
{
"decile": 1,
"n": 100,
"pred_mean": 0.28847,
"actual_mortality": 0.0
},
{
"decile": 2,
"n": 100,
"pred_mean": 0.33177,
"actual_mortality": 0.04
},
{
"decile": 3,
"n": 100,
"pred_mean": 0.36548,
"actual_mortality": 0.05
},
{
"decile": 4,
"n": 100,
"pred_mean": 0.4004,
"actual_mortality": 0.04
},
{
"decile": 5,
"n": 100,
"pred_mean": 0.43494,
"actual_mortality": 0.04
},
{
"decile": 6,
"n": 100,
"pred_mean": 0.47166,
"actual_mortality": 0.02
},
{
"decile": 7,
"n": 100,
"pred_mean": 0.51126,
"actual_mortality": 0.06
},
{
"decile": 8,
"n": 100,
"pred_mean": 0.5599,
"actual_mortality": 0.08
},
{
"decile": 9,
"n": 100,
"pred_mean": 0.65698,
"actual_mortality": 0.14
}
],
"auroc": 0.6895570728291316,
"auroc_ci_lo": 0.6165477435477436,
"auroc_ci_hi": 0.7575338334891509,
"auprc": 0.11894691697321907,
"auprc_ci_lo": 0.0761855785371724,
"auprc_ci_hi": 0.20169722298294696,
"brier": 0.19448950856037123,
"n_effective_bootstraps": 500,
"calibration_intercept": -2.7901140761804513,
"calibration_slope": 1.3055550649823606,
"o_over_e": 0.11338654179088853
},
"variants": [
{
"name": "v1_lgbm_baseline",
"label": "LightGBM (baseline, no brdav)",
"n_features": 26,
"uses_brdav": false,
"train_seconds": 2.33,
"calibration_curve": [
{
"decile": 0,
"n": 100,
"pred_mean": 1e-05,
"actual_mortality": 0.02
},
{
"decile": 1,
"n": 100,
"pred_mean": 3e-05,
"actual_mortality": 0.02
},
{
"decile": 2,
"n": 100,
"pred_mean": 8e-05,
"actual_mortality": 0.03
},
{
"decile": 3,
"n": 100,
"pred_mean": 0.00015,
"actual_mortality": 0.02
},
{
"decile": 4,
"n": 100,
"pred_mean": 0.0003,
"actual_mortality": 0.01
},
{
"decile": 5,
"n": 100,
"pred_mean": 0.00059,
"actual_mortality": 0.03
},
{
"decile": 6,
"n": 100,
"pred_mean": 0.00124,
"actual_mortality": 0.07
},
{
"decile": 7,
"n": 100,
"pred_mean": 0.00261,
"actual_mortality": 0.03
},
{
"decile": 8,
"n": 100,
"pred_mean": 0.00792,
"actual_mortality": 0.08
},
{
"decile": 9,
"n": 100,
"pred_mean": 0.15107,
"actual_mortality": 0.17
}
],
"calibration_raw_o_over_e": 2.92701601247577,
"auroc": 0.7431832107843136,
"auroc_ci_lo": 0.6654971932638332,
"auroc_ci_hi": 0.8125458184007442,
"auprc": 0.14339560395853382,
"auprc_ci_lo": 0.0857052924645232,
"auprc_ci_hi": 0.21995800062045834,
"brier": 0.04263326876605697,
"n_effective_bootstraps": 500,
"calibration_intercept": 0.0016152897089048443,
"calibration_slope": 1.0006689212138913,
"o_over_e": 0.9999999791666669
},
{
"name": "v2_lgbm_brdav",
"label": "LightGBM + brdav as feature 27",
"n_features": 27,
"uses_brdav": true,
"train_seconds": 2.16,
"calibration_curve": [
{
"decile": 0,
"n": 100,
"pred_mean": 1e-05,
"actual_mortality": 0.02
},
{
"decile": 1,
"n": 100,
"pred_mean": 3e-05,
"actual_mortality": 0.01
},
{
"decile": 2,
"n": 100,
"pred_mean": 7e-05,
"actual_mortality": 0.02
},
{
"decile": 3,
"n": 100,
"pred_mean": 0.00013,
"actual_mortality": 0.02
},
{
"decile": 4,
"n": 100,
"pred_mean": 0.00027,
"actual_mortality": 0.02
},
{
"decile": 5,
"n": 100,
"pred_mean": 0.00056,
"actual_mortality": 0.06
},
{
"decile": 6,
"n": 100,
"pred_mean": 0.00113,
"actual_mortality": 0.06
},
{
"decile": 7,
"n": 100,
"pred_mean": 0.00258,
"actual_mortality": 0.02
},
{
"decile": 8,
"n": 100,
"pred_mean": 0.00771,
"actual_mortality": 0.08
},
{
"decile": 9,
"n": 100,
"pred_mean": 0.1671,
"actual_mortality": 0.17
}
],
"calibration_raw_o_over_e": 2.6728012193710384,
"auroc": 0.7544314600840336,
"auroc_ci_lo": 0.6808597309472667,
"auroc_ci_hi": 0.8211254969449558,
"auprc": 0.16181000805974416,
"auprc_ci_lo": 0.09665686670526094,
"auprc_ci_hi": 0.25631319638298095,
"brier": 0.04219658993117402,
"n_effective_bootstraps": 500,
"calibration_intercept": 0.002757468732288981,
"calibration_slope": 1.0011796001615378,
"o_over_e": 0.9999999791666672
},
{
"name": "v3_stacking",
"label": "Stacking (LGBM + CatBoost + LR), brdav-aware",
"n_features": 27,
"uses_brdav": true,
"train_seconds": 11.18,
"calibration_curve": [
{
"decile": 0,
"n": 100,
"pred_mean": 0.02902,
"actual_mortality": 0.0
},
{
"decile": 1,
"n": 100,
"pred_mean": 0.02978,
"actual_mortality": 0.0
},
{
"decile": 2,
"n": 100,
"pred_mean": 0.03047,
"actual_mortality": 0.01
},
{
"decile": 3,
"n": 100,
"pred_mean": 0.03139,
"actual_mortality": 0.04
},
{
"decile": 4,
"n": 100,
"pred_mean": 0.03252,
"actual_mortality": 0.04
},
{
"decile": 5,
"n": 100,
"pred_mean": 0.03401,
"actual_mortality": 0.03
},
{
"decile": 6,
"n": 100,
"pred_mean": 0.03637,
"actual_mortality": 0.06
},
{
"decile": 7,
"n": 100,
"pred_mean": 0.04001,
"actual_mortality": 0.04
},
{
"decile": 8,
"n": 100,
"pred_mean": 0.04848,
"actual_mortality": 0.09
},
{
"decile": 9,
"n": 100,
"pred_mean": 0.14087,
"actual_mortality": 0.17
}
],
"calibration_raw_o_over_e": 1.059792826178591,
"auroc": 0.7805278361344538,
"auroc_ci_lo": 0.7261439954175871,
"auroc_ci_hi": 0.8368027552892396,
"auprc": 0.16801679625463095,
"auprc_ci_lo": 0.10060214737859723,
"auprc_ci_hi": 0.2654968074897177,
"brier": 0.041887173527302694,
"n_effective_bootstraps": 500,
"calibration_intercept": 0.001785150948646234,
"calibration_slope": 1.000827044521521,
"o_over_e": 0.9999999791666673
},
{
"name": "v4_tabpfn",
"label": "TabPFN v2 (foundation model, brdav-aware)",
"n_features": 27,
"uses_brdav": true,
"train_seconds": 1.04,
"calibration_curve": [
{
"decile": 0,
"n": 100,
"pred_mean": 0.00343,
"actual_mortality": 0.0
},
{
"decile": 1,
"n": 100,
"pred_mean": 0.00521,
"actual_mortality": 0.0
},
{
"decile": 2,
"n": 100,
"pred_mean": 0.00688,
"actual_mortality": 0.0
},
{
"decile": 3,
"n": 100,
"pred_mean": 0.00903,
"actual_mortality": 0.02
},
{
"decile": 4,
"n": 100,
"pred_mean": 0.01149,
"actual_mortality": 0.03
},
{
"decile": 5,
"n": 101,
"pred_mean": 0.01558,
"actual_mortality": 0.0396
},
{
"decile": 6,
"n": 99,
"pred_mean": 0.02114,
"actual_mortality": 0.07071
},
{
"decile": 7,
"n": 100,
"pred_mean": 0.03155,
"actual_mortality": 0.04
},
{
"decile": 8,
"n": 100,
"pred_mean": 0.05292,
"actual_mortality": 0.08
},
{
"decile": 9,
"n": 100,
"pred_mean": 0.17746,
"actual_mortality": 0.2
}
],
"calibration_raw_o_over_e": 1.4344382871541639,
"auroc": 0.8143601190476191,
"auroc_ci_lo": 0.7706158022485099,
"auroc_ci_hi": 0.8619519432623508,
"auprc": 0.1682243795640331,
"auprc_ci_lo": 0.10780224461587291,
"auprc_ci_hi": 0.2620784350727932,
"brier": 0.041644217848314936,
"n_effective_bootstraps": 500,
"calibration_intercept": -0.013026845470976038,
"calibration_slope": 0.994541501238136,
"o_over_e": 0.9999765165840673
}
],
"external_benchmarks": [
{
"name": "Br\u00fcggemann 2024 Swin-UNETR",
"cohort": "n=1,449 real Z\u00fcrich TAVR cohort",
"auroc": "0.725",
"endpoint": "all-cause follow-up mortality",
"url": "https://www.nature.com/articles/s41598-024-63022-x"
},
{
"name": "Cui 2025 MULTINet",
"cohort": "n=761 real MIMIC-IV TAVR cohort",
"auroc": "~0.78",
"endpoint": "in-hospital mortality",
"url": "https://www.mdpi.com/2077-0383/14/24/8620"
},
{
"name": "STS-PROM 2018",
"cohort": "trained on n=141k real SAVR; reported on TAVI external",
"auroc": "0.62\u20130.66",
"endpoint": "30-day mortality",
"url": "https://riskcalc.sts.org"
},
{
"name": "ACC/STS TVT (Edwards 2016)",
"cohort": "n=13,718 real TVT registry",
"auroc": "0.66\u20130.68",
"endpoint": "30-day mortality",
"url": "https://tools.acc.org/tavrrisk"
},
{
"name": "EuroSCORE II",
"cohort": "n=22k real cardiac surgery",
"auroc": "0.62",
"endpoint": "30-day mortality",
"url": "http://www.euroscore.org"
},
{
"name": "TRIM (Heinze 2023)",
"cohort": "n=22,283 real GARY",
"auroc": "0.75 (Swiss external)",
"endpoint": "30-day mortality",
"url": "https://academic.oup.com/ehjdh"
}
]
}